diff --git a/.DS_Store b/.DS_Store new file mode 100644 index 0000000000000000000000000000000000000000..094ec55be48c5ee180d007729299385e91e16d32 Binary files /dev/null and b/.DS_Store differ diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000000000000000000000000000000000000..e5526a95c863688dd1b2310b5828e8f4c411df84 --- /dev/null +++ b/.gitattributes @@ -0,0 +1,62 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tar filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text +single_instrument/bass/HTDemucs4_FT_bass.th filter=lfs diff=lfs merge=lfs -text +single_instrument/drums_percussion/HTDemucs4_FT_drums.th filter=lfs diff=lfs merge=lfs -text +legacy_models/fa0cb7f9-100d8bf4.th filter=lfs diff=lfs merge=lfs -text +legacy_models/cfa93e08-61801ae1.th filter=lfs diff=lfs merge=lfs -text +legacy_models/75fc33f5-1941ce65.th filter=lfs diff=lfs merge=lfs -text +legacy_models/demucs48_hq-28a1282c.th filter=lfs diff=lfs merge=lfs -text +legacy_models/902315c2-b39ce9c9.th filter=lfs diff=lfs merge=lfs -text +legacy_models/demucs_extra-3646af93.th filter=lfs diff=lfs merge=lfs -text +legacy_models/7d865c68-3d5dd56b.th filter=lfs diff=lfs merge=lfs -text +legacy_models/demucs_extra.th filter=lfs diff=lfs merge=lfs -text +legacy_models/ebf34a2db.th filter=lfs diff=lfs merge=lfs -text +legacy_models/demucs_unittest-09ebc15f.th filter=lfs diff=lfs merge=lfs -text +legacy_models/demucs.th filter=lfs diff=lfs merge=lfs -text +legacy_models/9a6b4851-03af0aa6.th filter=lfs diff=lfs merge=lfs -text +legacy_models/light.th filter=lfs diff=lfs merge=lfs -text +legacy_models/HTDemucs4_FT_other.th filter=lfs diff=lfs merge=lfs -text +legacy_models/1ef250f1-592467ce.th filter=lfs diff=lfs merge=lfs -text +legacy_models/demucs-e07c671f.th filter=lfs diff=lfs merge=lfs -text +legacy_models/e51eebcc-c1b80bdd.th filter=lfs diff=lfs merge=lfs -text +legacy_models/tasnet-beb46fac.th filter=lfs diff=lfs merge=lfs -text +legacy_models/tasnet.th filter=lfs diff=lfs merge=lfs -text +legacy_models/7ecf8ec1-70f50cc9.th filter=lfs diff=lfs merge=lfs -text +legacy_models/0d19c1c6-0f06f20e.th filter=lfs diff=lfs merge=lfs -text +vocal/vocal_extraction/HTDemucs4_FT_vocals_official.th filter=lfs diff=lfs merge=lfs -text +music_stems/drum_stems/model_drumsep.th filter=lfs diff=lfs merge=lfs -text +music_stems/four_stem/HTDemucs4.th filter=lfs diff=lfs merge=lfs -text +music_stems/six_stem/HTDemucs4_6stems.th filter=lfs diff=lfs merge=lfs -text diff --git a/.msc b/.msc new file mode 100644 index 0000000000000000000000000000000000000000..526a9f679bd9554fe16add1f8c30ae47da48e081 Binary files /dev/null and b/.msc differ diff --git a/.mv b/.mv new file mode 100644 index 0000000000000000000000000000000000000000..1c908be2fd04516d925f616731bdecf25e7ef5cb --- /dev/null +++ b/.mv @@ -0,0 +1 @@ +Revision:master,CreatedAt:1779378740 \ No newline at end of file diff --git a/INDEX.tsv b/INDEX.tsv new file mode 100644 index 0000000000000000000000000000000000000000..960cd68d322ec5c156394ea6ffaf923f4249456f --- /dev/null +++ b/INDEX.tsv @@ -0,0 +1,366 @@ +source_name size relpath source +cleanup 66541845 cleanup/audio_restoration/Apollo_LQ_MP3_restoration.ckpt /Volumes/2T/pymss/all_models/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.ckpt +cleanup 66528927 cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.pt /Volumes/2T/pymss/all_models/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.pt +cleanup 66532080 cleanup/audio_restoration/apollo_model.ckpt /Volumes/2T/pymss/all_models/cleanup/audio_restoration/apollo_model.ckpt +cleanup 38159792 cleanup/audio_restoration/apollo_model_v2.ckpt /Volumes/2T/pymss/all_models/cleanup/audio_restoration/apollo_model_v2.ckpt +cleanup 615867827 cleanup/audio_restoration/epoch=54-val_loss=-17.6221.ckpt /Volumes/2T/pymss/all_models/cleanup/audio_restoration/epoch=54-val_loss=-17.6221.ckpt +cleanup 194278526 cleanup/audio_restoration/model_apollo_vocals_ep_54.ckpt /Volumes/2T/pymss/all_models/cleanup/audio_restoration/model_apollo_vocals_ep_54.ckpt +cleanup 913096801 cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt /Volumes/2T/pymss/all_models/cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt +cleanup 913097300 cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt +cleanup 913097300 cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt +cleanup 395602493 cleanup/noise_bleed_control/huge_scnet_4stems_bleedless.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/huge_scnet_4stems_bleedless.ckpt +cleanup 913101368 cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.ckpt +cleanup 913102724 cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.ckpt +cleanup 913026650 cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.ckpt +cleanup 913026650 cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt +cleanup 913026650 cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt +cleanup 913026650 cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt /Volumes/2T/pymss/all_models/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt +experimental_general 740263433 experimental_general/BS-Roformer_LargeV1.ckpt /Volumes/2T/pymss/all_models/experimental_general/BS-Roformer_LargeV1.ckpt +experimental_general 613995746 experimental_general/BS_Roformer_FNF_V1.ckpt /Volumes/2T/pymss/all_models/experimental_general/BS_Roformer_FNF_V1.ckpt +experimental_general 913091027 experimental_general/Fullness.ckpt /Volumes/2T/pymss/all_models/experimental_general/Fullness.ckpt +experimental_general 913026650 experimental_general/INSTV8.ckpt /Volumes/2T/pymss/all_models/experimental_general/INSTV8.ckpt +experimental_general 913026650 experimental_general/INSTV9.ckpt /Volumes/2T/pymss/all_models/experimental_general/INSTV9.ckpt +experimental_general 183379219 experimental_general/MDX23C_D1581.ckpt /Volumes/2T/pymss/all_models/experimental_general/MDX23C_D1581.ckpt +experimental_general 1479738496 experimental_general/MelBandRoformerBigSYHFTV1.ckpt /Volumes/2T/pymss/all_models/experimental_general/MelBandRoformerBigSYHFTV1.ckpt +experimental_general 913096702 experimental_general/MelBandRoformerSYHFT.ckpt /Volumes/2T/pymss/all_models/experimental_general/MelBandRoformerSYHFT.ckpt +experimental_general 913106158 experimental_general/MelBandRoformerSYHFTB1.ckpt /Volumes/2T/pymss/all_models/experimental_general/MelBandRoformerSYHFTB1.ckpt +experimental_general 913090472 experimental_general/MelBandRoformerSYHFTV2.5.ckpt /Volumes/2T/pymss/all_models/experimental_general/MelBandRoformerSYHFTV2.5.ckpt +experimental_general 913095346 experimental_general/MelBandRoformerSYHFTV2.ckpt /Volumes/2T/pymss/all_models/experimental_general/MelBandRoformerSYHFTV2.ckpt +experimental_general 913090472 experimental_general/MelBandRoformerSYHFTV3Epsilon.ckpt /Volumes/2T/pymss/all_models/experimental_general/MelBandRoformerSYHFTV3Epsilon.ckpt +experimental_general 146867952 experimental_general/apollo_model_uni.ckpt /Volumes/2T/pymss/all_models/experimental_general/apollo_model_uni.ckpt +experimental_general 204682749 experimental_general/bs_exp_siameseroformer.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_exp_siameseroformer.ckpt +experimental_general 274890855 experimental_general/bs_hyperace.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_hyperace.ckpt +experimental_general 204485563 experimental_general/bs_roformer_anvuew_sdr_12.45.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_roformer_anvuew_sdr_12.45.ckpt +experimental_general 204493312 experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.ckpt +experimental_general 204502577 experimental_general/bs_roformer_mag_anvuew.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_roformer_mag_anvuew.ckpt +experimental_general 639326600 experimental_general/bs_roformer_revive.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_roformer_revive.ckpt +experimental_general 639326600 experimental_general/bs_roformer_revive2.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_roformer_revive2.ckpt +experimental_general 639326600 experimental_general/bs_roformer_revive3e.ckpt /Volumes/2T/pymss/all_models/experimental_general/bs_roformer_revive3e.ckpt +experimental_general 149097604 experimental_general/checkpoint-cmn_state_dict.ckpt /Volumes/2T/pymss/all_models/experimental_general/checkpoint-cmn_state_dict.ckpt +experimental_general 149097604 experimental_general/checkpoint-deu_state_dict.ckpt /Volumes/2T/pymss/all_models/experimental_general/checkpoint-deu_state_dict.ckpt +experimental_general 149097604 experimental_general/checkpoint-eng_state_dict.ckpt /Volumes/2T/pymss/all_models/experimental_general/checkpoint-eng_state_dict.ckpt +experimental_general 149097604 experimental_general/checkpoint-fao_state_dict.ckpt /Volumes/2T/pymss/all_models/experimental_general/checkpoint-fao_state_dict.ckpt +experimental_general 149097604 experimental_general/checkpoint-fra_state_dict.ckpt /Volumes/2T/pymss/all_models/experimental_general/checkpoint-fra_state_dict.ckpt +experimental_general 149097604 experimental_general/checkpoint-spa_state_dict.ckpt /Volumes/2T/pymss/all_models/experimental_general/checkpoint-spa_state_dict.ckpt +experimental_general 913837877 experimental_general/mbr-win10-sink8.ckpt /Volumes/2T/pymss/all_models/experimental_general/mbr-win10-sink8.ckpt +experimental_general 202573672 experimental_general/mel_band_roformer_small.ckpt /Volumes/2T/pymss/all_models/experimental_general/mel_band_roformer_small.ckpt +experimental_general 202583890 experimental_general/melband_roformer_small_v1.ckpt /Volumes/2T/pymss/all_models/experimental_general/melband_roformer_small_v1.ckpt +experimental_general 913036186 experimental_general/merge_syhft.ckpt /Volumes/2T/pymss/all_models/experimental_general/merge_syhft.ckpt +experimental_general 94204810 experimental_general/mini-bs-roformer-v2-46.8M.ckpt /Volumes/2T/pymss/all_models/experimental_general/mini-bs-roformer-v2-46.8M.ckpt +experimental_general 913106158 experimental_general/model2.ckpt /Volumes/2T/pymss/all_models/experimental_general/model2.ckpt +experimental_general 913106158 experimental_general/model3.ckpt /Volumes/2T/pymss/all_models/experimental_general/model3.ckpt +experimental_general 393068365 experimental_general/model_bs_roformer_ep_937_sdr_10.5309.ckpt /Volumes/2T/pymss/all_models/experimental_general/model_bs_roformer_ep_937_sdr_10.5309.ckpt +experimental_general 42458509 experimental_general/model_jazz_scnet.ckpt /Volumes/2T/pymss/all_models/experimental_general/model_jazz_scnet.ckpt +experimental_general 168869389 experimental_general/model_jazz_scnet_large.ckpt /Volumes/2T/pymss/all_models/experimental_general/model_jazz_scnet_large.ckpt +experimental_general 216190397 experimental_general/model_jazz_scnet_xl.ckpt /Volumes/2T/pymss/all_models/experimental_general/model_jazz_scnet_xl.ckpt +experimental_general 42434765 experimental_general/model_jazz_split_scnet.ckpt /Volumes/2T/pymss/all_models/experimental_general/model_jazz_split_scnet.ckpt +experimental_general 2739398173 experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt /Volumes/2T/pymss/all_models/experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt +experimental_general 4645148401 experimental_general/model_melband_roformer_ep_124_si_sdr_increase_7.6621.ckpt /Volumes/2T/pymss/all_models/experimental_general/model_melband_roformer_ep_124_si_sdr_increase_7.6621.ckpt +instrumental 913026650 instrumental/Inst_ExperimentalV1.ckpt /Volumes/2T/pymss/all_models/instrumental/Inst_ExperimentalV1.ckpt +instrumental 913027055 instrumental/Inst_Fv8.ckpt /Volumes/2T/pymss/all_models/instrumental/Inst_Fv8.ckpt +instrumental 913090472 instrumental/Inst_Fv9.ckpt /Volumes/2T/pymss/all_models/instrumental/Inst_Fv9.ckpt +instrumental 913091027 instrumental/Inst_GaboxFv9.ckpt /Volumes/2T/pymss/all_models/instrumental/Inst_GaboxFv9.ckpt +instrumental 913026650 instrumental/Inst_GaboxV7.ckpt /Volumes/2T/pymss/all_models/instrumental/Inst_GaboxV7.ckpt +instrumental 913090472 instrumental/Neo_InstVFX.ckpt /Volumes/2T/pymss/all_models/instrumental/Neo_InstVFX.ckpt +instrumental 238214371 instrumental/bs_large_v2_inst.ckpt /Volumes/2T/pymss/all_models/instrumental/bs_large_v2_inst.ckpt +instrumental 288725646 instrumental/bs_roformer_inst_hyperacev2.ckpt /Volumes/2T/pymss/all_models/instrumental/bs_roformer_inst_hyperacev2.ckpt +instrumental 913106900 instrumental/essid_mel_inst.ckpt /Volumes/2T/pymss/all_models/instrumental/essid_mel_inst.ckpt +instrumental 913107578 instrumental/essid_mel_inst_old.ckpt /Volumes/2T/pymss/all_models/instrumental/essid_mel_inst_old.ckpt +instrumental 489571079 instrumental/inst_gaboxFlowersV10.ckpt /Volumes/2T/pymss/all_models/instrumental/inst_gaboxFlowersV10.ckpt +instrumental 2739327117 instrumental/last_mel_band_roformer.ckpt /Volumes/2T/pymss/all_models/instrumental/last_mel_band_roformer.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_2_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_2_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_3_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_3_gabox.ckpt +instrumental 913106900 instrumental/mel_band_roformer_instrumental_becruily.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_becruily.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt +instrumental 913091027 instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt +instrumental 913031195 instrumental/mel_band_roformer_instrumental_fv7z_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fv7z_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_fv8_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fv8_gabox.ckpt +instrumental 913027055 instrumental/mel_band_roformer_instrumental_fv8b_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fv8b_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_fvx_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_fvx_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv5_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv5_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv5n_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv5n_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv6_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv6_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv6n_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv6n_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv7_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv7_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv7n_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv7n_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv8_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv8_gabox.ckpt +instrumental 913026650 instrumental/mel_band_roformer_instrumental_instv8n_gabox.ckpt /Volumes/2T/pymss/all_models/instrumental/mel_band_roformer_instrumental_instv8n_gabox.ckpt +instrumental 2739362141 instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt /Volumes/2T/pymss/all_models/instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt +instrumental 913097300 instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt /Volumes/2T/pymss/all_models/instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt +instrumental 202573672 instrumental/small_inst.ckpt /Volumes/2T/pymss/all_models/instrumental/small_inst.ckpt +karaoke 913090472 karaoke/Karaoke_GaboxV1.ckpt /Volumes/2T/pymss/all_models/karaoke/Karaoke_GaboxV1.ckpt +karaoke 204483448 karaoke/bs_karaoke_gabox_IS.ckpt /Volumes/2T/pymss/all_models/karaoke/bs_karaoke_gabox_IS.ckpt +karaoke 204486925 karaoke/bs_roformer_karaoke_anvuew.ckpt /Volumes/2T/pymss/all_models/karaoke/bs_roformer_karaoke_anvuew.ckpt +karaoke 204436907 karaoke/bs_roformer_karaoke_frazer_becruily.ckpt /Volumes/2T/pymss/all_models/karaoke/bs_roformer_karaoke_frazer_becruily.ckpt +karaoke 913091027 karaoke/mel_band_karaoke_fusion_aggressive.ckpt /Volumes/2T/pymss/all_models/karaoke/mel_band_karaoke_fusion_aggressive.ckpt +karaoke 913091027 karaoke/mel_band_karaoke_fusion_standard.ckpt /Volumes/2T/pymss/all_models/karaoke/mel_band_karaoke_fusion_standard.ckpt +karaoke 913090472 karaoke/mel_band_roformer_bve_gonza.ckpt /Volumes/2T/pymss/all_models/karaoke/mel_band_roformer_bve_gonza.ckpt +karaoke 1719139254 karaoke/mel_band_roformer_karaoke_becruily.ckpt /Volumes/2T/pymss/all_models/karaoke/mel_band_roformer_karaoke_becruily.ckpt +karaoke 913026650 karaoke/mel_band_roformer_karaoke_gabox.ckpt /Volumes/2T/pymss/all_models/karaoke/mel_band_roformer_karaoke_gabox.ckpt +karaoke 913090472 karaoke/mel_band_roformer_karaoke_gabox_v2.ckpt /Volumes/2T/pymss/all_models/karaoke/mel_band_roformer_karaoke_gabox_v2.ckpt +karaoke 913096801 karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt /Volumes/2T/pymss/all_models/karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt +karaoke 202574227 karaoke/small_karaoke_gaboxaufr.ckpt /Volumes/2T/pymss/all_models/karaoke/small_karaoke_gaboxaufr.ckpt +legacy_models 178048329 legacy_models/0d19c1c6-0f06f20e.th /Volumes/2T/pymss/all_models/legacy_models/0d19c1c6-0f06f20e.th +legacy_models 267602121 legacy_models/1ef250f1-592467ce.th /Volumes/2T/pymss/all_models/legacy_models/1ef250f1-592467ce.th +legacy_models 167407275 legacy_models/75fc33f5-1941ce65.th /Volumes/2T/pymss/all_models/legacy_models/75fc33f5-1941ce65.th +legacy_models 167918783 legacy_models/7d865c68-3d5dd56b.th /Volumes/2T/pymss/all_models/legacy_models/7d865c68-3d5dd56b.th +legacy_models 178048329 legacy_models/7ecf8ec1-70f50cc9.th /Volumes/2T/pymss/all_models/legacy_models/7ecf8ec1-70f50cc9.th +legacy_models 167405611 legacy_models/902315c2-b39ce9c9.th /Volumes/2T/pymss/all_models/legacy_models/902315c2-b39ce9c9.th +legacy_models 267602121 legacy_models/9a6b4851-03af0aa6.th /Volumes/2T/pymss/all_models/legacy_models/9a6b4851-03af0aa6.th +legacy_models 84141271 legacy_models/HTDemucs4_FT_other.th /Volumes/2T/pymss/all_models/legacy_models/HTDemucs4_FT_other.th +legacy_models 167399275 legacy_models/cfa93e08-61801ae1.th /Volumes/2T/pymss/all_models/legacy_models/cfa93e08-61801ae1.th +legacy_models 1062738817 legacy_models/demucs-e07c671f.th /Volumes/2T/pymss/all_models/legacy_models/demucs-e07c671f.th +legacy_models 2594202563 legacy_models/demucs.th /Volumes/2T/pymss/all_models/legacy_models/demucs.th +legacy_models 597870849 legacy_models/demucs48_hq-28a1282c.th /Volumes/2T/pymss/all_models/legacy_models/demucs48_hq-28a1282c.th +legacy_models 1062738817 legacy_models/demucs_extra-3646af93.th /Volumes/2T/pymss/all_models/legacy_models/demucs_extra-3646af93.th +legacy_models 2594202563 legacy_models/demucs_extra.th /Volumes/2T/pymss/all_models/legacy_models/demucs_extra.th +legacy_models 4194049 legacy_models/demucs_unittest-09ebc15f.th /Volumes/2T/pymss/all_models/legacy_models/demucs_unittest-09ebc15f.th +legacy_models 167399275 legacy_models/e51eebcc-c1b80bdd.th /Volumes/2T/pymss/all_models/legacy_models/e51eebcc-c1b80bdd.th +legacy_models 167390507 legacy_models/ebf34a2db.th /Volumes/2T/pymss/all_models/legacy_models/ebf34a2db.th +legacy_models 167405675 legacy_models/fa0cb7f9-100d8bf4.th /Volumes/2T/pymss/all_models/legacy_models/fa0cb7f9-100d8bf4.th +legacy_models 1062727712 legacy_models/light.th /Volumes/2T/pymss/all_models/legacy_models/light.th +legacy_models 44035981 legacy_models/tasnet-beb46fac.th /Volumes/2T/pymss/all_models/legacy_models/tasnet-beb46fac.th +legacy_models 43973949 legacy_models/tasnet.th /Volumes/2T/pymss/all_models/legacy_models/tasnet.th +legacy_vr 223650277 legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.pth +legacy_vr 127139365 legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.pth +legacy_vr 127139365 legacy_vr/vr_deecho/UVR-De-Echo-Normal.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_deecho/UVR-De-Echo-Normal.pth +legacy_vr 223650277 legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.pth +legacy_vr 17922277 legacy_vr/vr_denoise/UVR-DeNoise-Lite.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_denoise/UVR-DeNoise-Lite.pth +legacy_vr 127139365 legacy_vr/vr_denoise/UVR-DeNoise.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_denoise/UVR-DeNoise.pth +legacy_vr 58928133 legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth +legacy_vr 59320122 legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.pth +legacy_vr 550120123 legacy_vr/vr_hp2/7_HP2-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_hp2/7_HP2-UVR.pth +legacy_vr 550131079 legacy_vr/vr_hp2/8_HP2-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_hp2/8_HP2-UVR.pth +legacy_vr 550131079 legacy_vr/vr_hp2/9_HP2-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_hp2/9_HP2-UVR.pth +legacy_vr 126792647 legacy_vr/vr_instrumental_extraction/1_HP-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_instrumental_extraction/1_HP-UVR.pth +legacy_vr 126782699 legacy_vr/vr_instrumental_extraction/2_HP-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_instrumental_extraction/2_HP-UVR.pth +legacy_vr 126782699 legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.pth +legacy_vr 126782699 legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.pth +legacy_vr 31861529 legacy_vr/vr_mgm/MGM_HIGHEND_v4.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_mgm/MGM_HIGHEND_v4.pth +legacy_vr 31938951 legacy_vr/vr_mgm/MGM_LOWEND_A_v4.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_mgm/MGM_LOWEND_A_v4.pth +legacy_vr 31938951 legacy_vr/vr_mgm/MGM_LOWEND_B_v4.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_mgm/MGM_LOWEND_B_v4.pth +legacy_vr 31938951 legacy_vr/vr_mgm/MGM_MAIN_v4.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_mgm/MGM_MAIN_v4.pth +legacy_vr 31938951 legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.pth +legacy_vr 31938951 legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.pth +legacy_vr 31938951 legacy_vr/vr_sp/12_SP-UVR-3B-44100.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_sp/12_SP-UVR-3B-44100.pth +legacy_vr 31938951 legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.pth +legacy_vr 31938951 legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.pth +legacy_vr 31938951 legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.pth +legacy_vr 31938951 legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.pth +legacy_vr 126792647 legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.pth +legacy_vr 126792647 legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.pth +legacy_vr 223661285 legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.pth /Volumes/2T/pymss/all_models/legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.pth +music_stems 149133378 music_stems/dnr_stems/checkpoint-multi_fixed.ckpt /Volumes/2T/pymss/all_models/music_stems/dnr_stems/checkpoint-multi_fixed.ckpt +music_stems 148891175 music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.ckpt /Volumes/2T/pymss/all_models/music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.ckpt +music_stems 437652699 music_stems/drum_stems/aufr33-jarredou_DrumSep_model_mdx23c_ep_141_sdr_10.8059.ckpt /Volumes/2T/pymss/all_models/music_stems/drum_stems/aufr33-jarredou_DrumSep_model_mdx23c_ep_141_sdr_10.8059.ckpt +music_stems 437648758 music_stems/drum_stems/drumsep_5stems_mdx23c_jarredou.ckpt /Volumes/2T/pymss/all_models/music_stems/drum_stems/drumsep_5stems_mdx23c_jarredou.ckpt +music_stems 303490678 music_stems/drum_stems/gilliaan_drumsV1.ckpt /Volumes/2T/pymss/all_models/music_stems/drum_stems/gilliaan_drumsV1.ckpt +music_stems 167400043 music_stems/drum_stems/model_drumsep.th /Volumes/2T/pymss/all_models/music_stems/drum_stems/model_drumsep.th +music_stems 84141911 music_stems/four_stem/HTDemucs4.th /Volumes/2T/pymss/all_models/music_stems/four_stem/HTDemucs4.th +music_stems 3764030874 music_stems/four_stem/MelBandRoformer4StemFTLarge.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/MelBandRoformer4StemFTLarge.ckpt +music_stems 168852258 music_stems/four_stem/SCNet-large_starrytong_fixed.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/SCNet-large_starrytong_fixed.ckpt +music_stems 527245586 music_stems/four_stem/bs_roformer_4stems_ft.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/bs_roformer_4stems_ft.ckpt +music_stems 149464276 music_stems/four_stem/fused_model_bs_conformer_sdr_9.18.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/fused_model_bs_conformer_sdr_9.18.ckpt +music_stems 395597693 music_stems/four_stem/huge_scnet_4stems_fullness.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/huge_scnet_4stems_fullness.ckpt +music_stems 395598013 music_stems/four_stem/huge_scnet_4stems_strong_fullness.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/huge_scnet_4stems_strong_fullness.ckpt +music_stems 395599229 music_stems/four_stem/huge_scnet_4stems_v1.2.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/huge_scnet_4stems_v1.2.ckpt +music_stems 3764030874 music_stems/four_stem/mel_band_roformer_4stems_large_ver1.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/mel_band_roformer_4stems_large_ver1.ckpt +music_stems 6411087858 music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.ckpt +music_stems 142774366 music_stems/four_stem/model_bs_mamba2_ep_11_sdr_6.8723.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_bs_mamba2_ep_11_sdr_6.8723.ckpt +music_stems 527385512 music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.ckpt +music_stems 448119834 music_stems/four_stem/model_mdx23c_ep_168_sdr_7.0207.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mdx23c_ep_168_sdr_7.0207.ckpt +music_stems 250899372 music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt +music_stems 234032732 music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt +music_stems 81867982 music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt +music_stems 81866140 music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt +music_stems 81890712 music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt +music_stems 118636682 music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt +music_stems 118634392 music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt +music_stems 29889628 music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt +music_stems 177475814 music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt +music_stems 214063778 music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.ckpt +music_stems 216189106 music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.ckpt +music_stems 214143287 music_stems/four_stem/model_scnet_masked_ep_111_sdr_9.8286.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_scnet_masked_ep_111_sdr_9.8286.ckpt +music_stems 42517058 music_stems/four_stem/model_scnet_masked_ep_156_sdr_8.8149.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_scnet_masked_ep_156_sdr_8.8149.ckpt +music_stems 168868194 music_stems/four_stem/model_scnet_sdr_9.3244.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_scnet_sdr_9.3244.ckpt +music_stems 41696502 music_stems/four_stem/model_scnet_tran_sdr_8.9272.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/model_scnet_tran_sdr_8.9272.ckpt +music_stems 42434986 music_stems/four_stem/scnet_checkpoint_musdb18.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/scnet_checkpoint_musdb18.ckpt +music_stems 3764030874 music_stems/four_stem/ver2.ckpt /Volumes/2T/pymss/all_models/music_stems/four_stem/ver2.ckpt +music_stems 77619808 music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.ckpt +music_stems 77624038 music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.ckpt +music_stems 77620513 music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.ckpt +music_stems 77618398 music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.ckpt +music_stems 77622628 music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.ckpt +music_stems 77619103 music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.ckpt +music_stems 77617693 music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.ckpt +music_stems 77622628 music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.ckpt +music_stems 77621218 music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.ckpt +music_stems 77624038 music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.ckpt +music_stems 77621218 music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.ckpt +music_stems 77621923 music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.ckpt +music_stems 77617693 music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.ckpt +music_stems 77619808 music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.ckpt +music_stems 77621218 music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.ckpt +music_stems 77614809 music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.ckpt +music_stems 77620513 music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.ckpt +music_stems 77619103 music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.ckpt +music_stems 77618398 music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.ckpt +music_stems 77620513 music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.ckpt +music_stems 77619808 music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.ckpt +music_stems 77618398 music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.ckpt +music_stems 77620513 music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.ckpt +music_stems 77618398 music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.ckpt +music_stems 77619103 music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.ckpt +music_stems 77619103 music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.ckpt +music_stems 77618398 music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.ckpt +music_stems 77618398 music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.ckpt +music_stems 77617693 music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.ckpt +music_stems 77616988 music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.ckpt +music_stems 77621218 music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.ckpt +music_stems 77616219 music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.ckpt +music_stems 77619103 music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.ckpt +music_stems 1368919887 music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt /Volumes/2T/pymss/all_models/music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt +music_stems 699412152 music_stems/six_stem/BS-Roformer-SW.ckpt /Volumes/2T/pymss/all_models/music_stems/six_stem/BS-Roformer-SW.ckpt +music_stems 54996327 music_stems/six_stem/HTDemucs4_6stems.th /Volumes/2T/pymss/all_models/music_stems/six_stem/HTDemucs4_6stems.th +music_stems 913026650 music_stems/six_stem/gabox_karaoke_25_02_2025.ckpt /Volumes/2T/pymss/all_models/music_stems/six_stem/gabox_karaoke_25_02_2025.ckpt +music_stems 699446752 music_stems/six_stem/logic_bs_roformer.ckpt /Volumes/2T/pymss/all_models/music_stems/six_stem/logic_bs_roformer.ckpt +reverb_echo_control 913031195 reverb_echo_control/dereverb/Lead_VocalDereverb.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/Lead_VocalDereverb.ckpt +reverb_echo_control 455864012 reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.ckpt +reverb_echo_control 204496717 reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt +reverb_echo_control 448098867 reverb_echo_control/dereverb/dereverb_mdx23c_sdr_6.9096.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/dereverb_mdx23c_sdr_6.9096.ckpt +reverb_echo_control 913107578 reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt +reverb_echo_control 913107578 reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt +reverb_echo_control 913097978 reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.ckpt +reverb_echo_control 118128452 reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.ckpt +reverb_echo_control 455864012 reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.ckpt +reverb_echo_control 170770820 reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.ckpt +reverb_echo_control 361499604 reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.ckpt +reverb_echo_control 438988833 reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.ckpt +reverb_echo_control 1708522165 reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.ckpt +reverb_echo_control 489578757 reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt +reverb_echo_control 835997896 reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt +reverb_echo_control 455862568 reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt +reverb_echo_control 455776577 reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt /Volumes/2T/pymss/all_models/reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt +single_instrument 84141271 single_instrument/bass/HTDemucs4_FT_bass.th /Volumes/2T/pymss/all_models/single_instrument/bass/HTDemucs4_FT_bass.th +single_instrument 42458509 single_instrument/bass/model_jazz_scnet_bass_drums.ckpt /Volumes/2T/pymss/all_models/single_instrument/bass/model_jazz_scnet_bass_drums.ckpt +single_instrument 84141271 single_instrument/drums_percussion/HTDemucs4_FT_drums.th /Volumes/2T/pymss/all_models/single_instrument/drums_percussion/HTDemucs4_FT_drums.th +single_instrument 42458509 single_instrument/drums_percussion/model_jazz_scnet_piano_drums.ckpt /Volumes/2T/pymss/all_models/single_instrument/drums_percussion/model_jazz_scnet_piano_drums.ckpt +single_instrument 45142183 single_instrument/guitar/becruily_guitar.ckpt /Volumes/2T/pymss/all_models/single_instrument/guitar/becruily_guitar.ckpt +single_instrument 276726667 single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt /Volumes/2T/pymss/all_models/single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt +single_instrument 303490678 single_instrument/strings/gilliaan_bowedstrings_bs_v1.ckpt /Volumes/2T/pymss/all_models/single_instrument/strings/gilliaan_bowedstrings_bs_v1.ckpt +spatial 869726129 spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.ckpt /Volumes/2T/pymss/all_models/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.ckpt +spatial 869726129 spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.ckpt /Volumes/2T/pymss/all_models/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.ckpt +spatial 204495806 spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.ckpt /Volumes/2T/pymss/all_models/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.ckpt +spatial 204435682 spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.ckpt /Volumes/2T/pymss/all_models/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.ckpt +spatial 437617081 spatial/phantom_center_wide/gilliaan_mdx23c_centerwide_v1.ckpt /Volumes/2T/pymss/all_models/spatial/phantom_center_wide/gilliaan_mdx23c_centerwide_v1.ckpt +spatial 214021674 spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.ckpt /Volumes/2T/pymss/all_models/spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.ckpt +special 168122809 special/model_htdemucs_ep_21_sdr_13.6970.ckpt /Volumes/2T/pymss/all_models/special/model_htdemucs_ep_21_sdr_13.6970.ckpt +special 437613512 special/model_mdx23c_ep_271_l1_freq_72.2383.ckpt /Volumes/2T/pymss/all_models/special/model_mdx23c_ep_271_l1_freq_72.2383.ckpt +special 42422546 special/model_scnet_ep_102_sdr_12.8941.ckpt /Volumes/2T/pymss/all_models/special/model_scnet_ep_102_sdr_12.8941.ckpt +special 42422170 special/model_scnet_ep_30_sdr_15.1291.ckpt /Volumes/2T/pymss/all_models/special/model_scnet_ep_30_sdr_15.1291.ckpt +vocal 835982664 vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt /Volumes/2T/pymss/all_models/vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt +vocal 835983746 vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.ckpt /Volumes/2T/pymss/all_models/vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.ckpt +vocal 42457314 vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.ckpt /Volumes/2T/pymss/all_models/vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.ckpt +vocal 42457314 vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.ckpt /Volumes/2T/pymss/all_models/vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.ckpt +vocal 42522921 vocal/choir_vocal/model_scnet_masked_ep_14_sdr_6.3693.ckpt /Volumes/2T/pymss/all_models/vocal/choir_vocal/model_scnet_masked_ep_14_sdr_6.3693.ckpt +vocal 42522921 vocal/choir_vocal/model_scnet_masked_ep_14_sdr_6.9052.ckpt /Volumes/2T/pymss/all_models/vocal/choir_vocal/model_scnet_masked_ep_14_sdr_6.9052.ckpt +vocal 630204333 vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt /Volumes/2T/pymss/all_models/vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt +vocal 102387933 vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.ckpt /Volumes/2T/pymss/all_models/vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.ckpt +vocal 435091173 vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.ckpt /Volumes/2T/pymss/all_models/vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.ckpt +vocal 59960415 vocal/vocal_extraction/16k_Vocal_Lightweight_MelBandRoformer.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/16k_Vocal_Lightweight_MelBandRoformer.ckpt +vocal 204510749 vocal/vocal_extraction/BS-Roformer-Resurrection.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/BS-Roformer-Resurrection.ckpt +vocal 84141271 vocal/vocal_extraction/HTDemucs4_FT_vocals_official.th /Volumes/2T/pymss/all_models/vocal/vocal_extraction/HTDemucs4_FT_vocals_official.th +vocal 913106900 vocal/vocal_extraction/Kim_MelBandRoformer.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/Kim_MelBandRoformer.ckpt +vocal 1479749810 vocal/vocal_extraction/big_beta5e.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/big_beta5e.ckpt +vocal 1557078584 vocal/vocal_extraction/big_beta6.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/big_beta6.ckpt +vocal 1708527586 vocal/vocal_extraction/big_beta6x.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/big_beta6x.ckpt +vocal 944675923 vocal/vocal_extraction/big_beta7.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/big_beta7.ckpt +vocal 192610602 vocal/vocal_extraction/bs_mamba2_vocals.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/bs_mamba2_vocals.ckpt +vocal 288724430 vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt +vocal 639254584 vocal/vocal_extraction/bs_roformer_vocals_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/bs_roformer_vocals_gabox.ckpt +vocal 913030778 vocal/vocal_extraction/inst_fv7b.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/inst_fv7b.ckpt +vocal 913026650 vocal/vocal_extraction/instv7beta2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/instv7beta2.ckpt +vocal 913026650 vocal/vocal_extraction/instv7beta3.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/instv7beta3.ckpt +vocal 913031195 vocal/vocal_extraction/instv7plus.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/instv7plus.ckpt +vocal 913100690 vocal/vocal_extraction/kimmel_unwa_ft.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/kimmel_unwa_ft.ckpt +vocal 913100690 vocal/vocal_extraction/kimmel_unwa_ft2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/kimmel_unwa_ft2.ckpt +vocal 913090472 vocal/vocal_extraction/kimmel_unwa_ft3_prev.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/kimmel_unwa_ft3_prev.ckpt +vocal 204483448 vocal/vocal_extraction/last_bs_roformer.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/last_bs_roformer.ckpt +vocal 913090472 vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.ckpt +vocal 913107578 vocal/vocal_extraction/mel_band_roformer_vocals_becruily.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_becruily.ckpt +vocal 913026650 vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.ckpt +vocal 913026650 vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.ckpt +vocal 913026650 vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.ckpt +vocal 913026650 vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.ckpt +vocal 913026650 vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.ckpt +vocal 913031195 vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.ckpt +vocal 913091027 vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.ckpt +vocal 913026650 vocal/vocal_extraction/mel_band_roformer_vocals_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_gabox.ckpt +vocal 913031195 vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.ckpt +vocal 1479745022 vocal/vocal_extraction/melband_roformer_big_beta1.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/melband_roformer_big_beta1.ckpt +vocal 1479745022 vocal/vocal_extraction/melband_roformer_big_beta2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/melband_roformer_big_beta2.ckpt +vocal 1479738496 vocal/vocal_extraction/melband_roformer_big_beta3.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/melband_roformer_big_beta3.ckpt +vocal 1574477088 vocal/vocal_extraction/melband_roformer_big_beta4.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/melband_roformer_big_beta4.ckpt +vocal 102410301 vocal/vocal_extraction/model_bs_polarformer_float16.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/model_bs_polarformer_float16.ckpt +vocal 639331213 vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.ckpt +vocal 639317465 vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.ckpt +vocal 639330302 vocal/vocal_extraction/model_bsrofomer_essid_vocals.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/model_bsrofomer_essid_vocals.ckpt +vocal 1007816988 vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt +vocal 725806842 vocal/vocal_extraction/model_vocals.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/model_vocals.ckpt +vocal 134917909 vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.ckpt +vocal 489571079 vocal/vocal_extraction/voc_fv7.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/voc_fv7.ckpt +vocal 913030778 vocal/vocal_extraction/vocfv7beta2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/vocfv7beta2.ckpt +vocal 913030778 vocal/vocal_extraction/vocfv7beta3.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_extraction/vocfv7beta3.ckpt +vocal 527119779 vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt +vocal 527121477 vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt +vocal 527121477 vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt +vocal 204483033 vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.ckpt +vocal 393351501 vocal/vocal_instrumental_dual/BS_Inst_EXP_VRL.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/BS_Inst_EXP_VRL.ckpt +vocal 913026650 vocal/vocal_instrumental_dual/INSTV10.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/INSTV10.ckpt +vocal 448093770 vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.ckpt +vocal 299892736 vocal/vocal_instrumental_dual/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa.ckpt +vocal 3407872 vocal/vocal_instrumental_dual/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa.ckpt +vocal 435006815 vocal/vocal_instrumental_dual/becruily_deux.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/becruily_deux.ckpt +vocal 332004435 vocal/vocal_instrumental_dual/bs_roformer_fno.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/bs_roformer_fno.ckpt +vocal 204483033 vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.ckpt +vocal 913043907 vocal/vocal_instrumental_dual/duality_v1.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/duality_v1.ckpt +vocal 913106900 vocal/vocal_instrumental_dual/essid_mel_vocal_v1.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/essid_mel_vocal_v1.ckpt +vocal 913106900 vocal/vocal_instrumental_dual/essid_mel_vocal_v2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/essid_mel_vocal_v2.ckpt +vocal 913090472 vocal/vocal_instrumental_dual/inst_v1_plus_test.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/inst_v1_plus_test.ckpt +vocal 913102724 vocal/vocal_instrumental_dual/inst_v1e.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/inst_v1e.ckpt +vocal 913090472 vocal/vocal_instrumental_dual/inst_v1e_plus.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/inst_v1e_plus.ckpt +vocal 913100690 vocal/vocal_instrumental_dual/melband_roformer_inst_v1.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/melband_roformer_inst_v1.ckpt +vocal 1574477088 vocal/vocal_instrumental_dual/melband_roformer_inst_v2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/melband_roformer_inst_v2.ckpt +vocal 1719116358 vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.ckpt +vocal 1719116358 vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.ckpt +vocal 944683368 vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.ckpt +vocal 448103606 vocal/vocal_instrumental_dual/model_mdx23c_ep_0_sdr_14.6632.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/model_mdx23c_ep_0_sdr_14.6632.ckpt +vocal 939903161 vocal/vocal_instrumental_dual/model_swin_upernet_ep_56_sdr_10.6703.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/model_swin_upernet_ep_56_sdr_10.6703.ckpt +vocal 168124336 vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.ckpt +vocal 448101203 vocal/vocal_instrumental_dual/model_vocals_mdx23c_sdr_10.17.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/model_vocals_mdx23c_sdr_10.17.ckpt +vocal 863683537 vocal/vocal_instrumental_dual/model_vocals_segm_models_sdr_9.77.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/model_vocals_segm_models_sdr_9.77.ckpt +vocal 204483448 vocal/vocal_instrumental_dual/neo_inst_beta.ckpt /Volumes/2T/pymss/all_models/vocal/vocal_instrumental_dual/neo_inst_beta.ckpt diff --git a/MANIFEST.json.before_duplicate_cleanup b/MANIFEST.json.before_duplicate_cleanup new file mode 100644 index 0000000000000000000000000000000000000000..79abbabc7fccbd1b0d374a0bff5e5b08470d6a9c --- /dev/null +++ b/MANIFEST.json.before_duplicate_cleanup @@ -0,0 +1,3555 @@ +{ + "generated_at": "2026-05-21 02:18:42", + "directory": "/Volumes/2T/pymss/all_models", + "method": "hardlinks where possible, symlinks fallback; original files retained", + "sources": [ + { + "name": "downloaded", + "root": "/Volumes/2T/pymss/benchmark_results/model_downloads" + }, + { + "name": "manual_models", + "root": "/Volumes/2T/pymss/models" + }, + { + "name": "msst_webui_pretrain", + "root": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain" + }, + { + "name": "pymss_pretrain", + "root": "/Volumes/2T/pymss/pretrain" + } + ], + "model_files": 440, + "linked_created": 21, + "already_existing": 419, + "skipped": [ + { + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__MelBand_BS_Roformers-pytorch/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial.ckpt", + "reason": "zero_size" + } + ], + "errors": [], + "items": [ + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/chenCFD__Guitar_Source_Separation/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/chenCFD__Guitar_Source_Separation/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt", + "relpath": "downloaded/github/chenCFD__Guitar_Source_Separation/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt", + "size": 276726667, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/jarredou__models/aufr33-jarredou_DrumSep_model_mdx23c_ep_141_sdr_10.8059.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/jarredou__models/aufr33-jarredou_DrumSep_model_mdx23c_ep_141_sdr_10.8059.ckpt", + "relpath": "downloaded/github/jarredou__models/aufr33-jarredou_DrumSep_model_mdx23c_ep_141_sdr_10.8059.ckpt", + "size": 437652699, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/jarredou__models/drumsep_5stems_mdx23c_jarredou.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/jarredou__models/drumsep_5stems_mdx23c_jarredou.ckpt", + "relpath": "downloaded/github/jarredou__models/drumsep_5stems_mdx23c_jarredou.ckpt", + "size": 437648758, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/bs_mamba2_vocals.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/bs_mamba2_vocals.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/bs_mamba2_vocals.ckpt", + "size": 192610602, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/fused_model_bs_conformer_sdr_9.18.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/fused_model_bs_conformer_sdr_9.18.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/fused_model_bs_conformer_sdr_9.18.ckpt", + "size": 149464276, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_bandit_plus_dnr_sdr_11.47.chpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_bandit_plus_dnr_sdr_11.47.chpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_bandit_plus_dnr_sdr_11.47.chpt", + "size": 148891175, + "suffix": ".chpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_bs_polarformer_float16.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_bs_polarformer_float16.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_bs_polarformer_float16.ckpt", + "size": 102410301, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_bs_roformer_ep_17_sdr_9.6568.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_bs_roformer_ep_17_sdr_9.6568.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_bs_roformer_ep_17_sdr_9.6568.ckpt", + "size": 527385512, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_drumsep.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_drumsep.th", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_drumsep.th", + "size": 167400043, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_mdx23c_ep_168_sdr_7.0207.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_mdx23c_ep_168_sdr_7.0207.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_mdx23c_ep_168_sdr_7.0207.ckpt", + "size": 448119834, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_mdx23c_ep_271_l1_freq_72.2383.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_mdx23c_ep_271_l1_freq_72.2383.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_mdx23c_ep_271_l1_freq_72.2383.ckpt", + "size": 437613512, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_ep_36_sdr_10.0891.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_ep_36_sdr_10.0891.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_ep_36_sdr_10.0891.ckpt", + "size": 214063778, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_ep_54_sdr_9.8051.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_ep_54_sdr_9.8051.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_ep_54_sdr_9.8051.ckpt", + "size": 216189106, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_masked_ep_111_sdr_9.8286.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_masked_ep_111_sdr_9.8286.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_masked_ep_111_sdr_9.8286.ckpt", + "size": 214143287, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_masked_ep_156_sdr_8.8149.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_masked_ep_156_sdr_8.8149.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_masked_ep_156_sdr_8.8149.ckpt", + "size": 42517058, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_sdr_9.3244.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_sdr_9.3244.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_sdr_9.3244.ckpt", + "size": 168868194, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_tran_sdr_8.9272.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_tran_sdr_8.9272.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_scnet_tran_sdr_8.9272.ckpt", + "size": 41696502, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "size": 939903161, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_htdemucs_sdr_8.78.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_htdemucs_sdr_8.78.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_htdemucs_sdr_8.78.ckpt", + "size": 168124336, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_mdx23c_sdr_10.17.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_mdx23c_sdr_10.17.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_mdx23c_sdr_10.17.ckpt", + "size": 448101203, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_segm_models_sdr_9.77.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_segm_models_sdr_9.77.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/model_vocals_segm_models_sdr_9.77.ckpt", + "size": 863683537, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.3/model_bandit_plus_dnr_sdr_11.47.chpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.3/model_bandit_plus_dnr_sdr_11.47.chpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.3/model_bandit_plus_dnr_sdr_11.47.chpt", + "size": 148891175, + "suffix": ".chpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.4/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.4/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.4/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt", + "size": 913096801, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.6/scnet_checkpoint_musdb18.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.6/scnet_checkpoint_musdb18.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.6/scnet_checkpoint_musdb18.ckpt", + "size": 42434986, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.7/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.7/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.7/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt", + "size": 913097300, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.7/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.7/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v.1.0.7/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt", + "size": 913097300, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_htdemucs_sdr_8.78.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_htdemucs_sdr_8.78.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_htdemucs_sdr_8.78.ckpt", + "size": 168124336, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_mdx23c_sdr_10.17.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_mdx23c_sdr_10.17.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_mdx23c_sdr_10.17.ckpt", + "size": 448101203, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_mel_band_roformer_sdr_8.42.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_mel_band_roformer_sdr_8.42.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_mel_band_roformer_sdr_8.42.ckpt", + "size": 134917909, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_segm_models_sdr_9.77.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_segm_models_sdr_9.77.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.0/model_vocals_segm_models_sdr_9.77.ckpt", + "size": 863683537, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.1/model_mdx23c_ep_168_sdr_7.0207.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.1/model_mdx23c_ep_168_sdr_7.0207.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.1/model_mdx23c_ep_168_sdr_7.0207.ckpt", + "size": 448119834, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.10/model_mdx23c_ep_271_l1_freq_72.2383.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.10/model_mdx23c_ep_271_l1_freq_72.2383.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.10/model_mdx23c_ep_271_l1_freq_72.2383.ckpt", + "size": 437613512, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt", + "size": 250899372, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt", + "size": 234032732, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt", + "size": 81867982, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt", + "size": 81866140, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt", + "size": 81890712, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt", + "size": 118636682, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt", + "size": 118634392, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt", + "size": 29889628, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.11/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt", + "size": 177475814, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.12/model_bs_roformer_ep_17_sdr_9.6568.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.12/model_bs_roformer_ep_17_sdr_9.6568.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.12/model_bs_roformer_ep_17_sdr_9.6568.ckpt", + "size": 527385512, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.13/model_scnet_ep_54_sdr_9.8051.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.13/model_scnet_ep_54_sdr_9.8051.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.13/model_scnet_ep_54_sdr_9.8051.ckpt", + "size": 216189106, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.14/model_scnet_tran_sdr_8.9272.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.14/model_scnet_tran_sdr_8.9272.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.14/model_scnet_tran_sdr_8.9272.ckpt", + "size": 41696502, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.15/model_scnet_ep_36_sdr_10.0891.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.15/model_scnet_ep_36_sdr_10.0891.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.15/model_scnet_ep_36_sdr_10.0891.ckpt", + "size": 214063778, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.16/model_scnet_masked_ep_156_sdr_8.8149.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.16/model_scnet_masked_ep_156_sdr_8.8149.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.16/model_scnet_masked_ep_156_sdr_8.8149.ckpt", + "size": 42517058, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.17/model_scnet_masked_ep_111_sdr_9.8286.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.17/model_scnet_masked_ep_111_sdr_9.8286.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.17/model_scnet_masked_ep_111_sdr_9.8286.ckpt", + "size": 214143287, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.18/fused_model_bs_conformer_sdr_9.18.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.18/fused_model_bs_conformer_sdr_9.18.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.18/fused_model_bs_conformer_sdr_9.18.ckpt", + "size": 149464276, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.19/bs_mamba2_vocals.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.19/bs_mamba2_vocals.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.19/bs_mamba2_vocals.ckpt", + "size": 192610602, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.2/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.2/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.2/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "size": 939903161, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.20/model_bs_polarformer_float16.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.20/model_bs_polarformer_float16.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.20/model_bs_polarformer_float16.ckpt", + "size": 102410301, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.21/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.21/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.21/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt", + "size": 1368919887, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.5/model_drumsep.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.5/model_drumsep.th", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.5/model_drumsep.th", + "size": 167400043, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.8/model_scnet_sdr_9.3244.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.8/model_scnet_sdr_9.3244.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.8/model_scnet_sdr_9.3244.ckpt", + "size": 168868194, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.9/model_bs_mamba2_ep_11_sdr_6.8723.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.9/model_bs_mamba2_ep_11_sdr_6.8723.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.9/model_bs_mamba2_ep_11_sdr_6.8723.ckpt", + "size": 142774366, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.9/SCNet-large_starrytong_fixed.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.9/SCNet-large_starrytong_fixed.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/releases/v1.0.9/SCNet-large_starrytong_fixed.ckpt", + "size": 168852258, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/SCNet-large_starrytong_fixed.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/SCNet-large_starrytong_fixed.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/SCNet-large_starrytong_fixed.ckpt", + "size": 168852258, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/github/ZFTurbo__Music-Source-Separation-Training/scnet_checkpoint_musdb18.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/github/ZFTurbo__Music-Source-Separation-Training/scnet_checkpoint_musdb18.ckpt", + "relpath": "downloaded/github/ZFTurbo__Music-Source-Separation-Training/scnet_checkpoint_musdb18.ckpt", + "size": 42434986, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/AliceN__Roformer-SpeechSep/speech_bs_roformer_AliceN_sdr_20.4440.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/AliceN__Roformer-SpeechSep/speech_bs_roformer_AliceN_sdr_20.4440.ckpt", + "relpath": "downloaded/huggingface/AliceN__Roformer-SpeechSep/speech_bs_roformer_AliceN_sdr_20.4440.ckpt", + "size": 102387933, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/AliceN__Roformer-SpeechSep/speech_mel_roformer_AliceN_sdr_16.5716.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/AliceN__Roformer-SpeechSep/speech_mel_roformer_AliceN_sdr_16.5716.ckpt", + "relpath": "downloaded/huggingface/AliceN__Roformer-SpeechSep/speech_mel_roformer_AliceN_sdr_16.5716.ckpt", + "size": 435091173, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_bleedless.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_bleedless.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_bleedless.ckpt", + "size": 395602493, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_fullness.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_fullness.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_fullness.ckpt", + "size": 395597693, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_strong_fullness.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_strong_fullness.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_strong_fullness.ckpt", + "size": 395598013, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_v1.2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_v1.2.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__Huge-SCNet-4stems/huge_scnet_4stems_v1.2.ckpt", + "size": 395599229, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__Mel-Band-Roformer_Duality/duality_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__Mel-Band-Roformer_Duality/duality_v1.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__Mel-Band-Roformer_Duality/duality_v1.ckpt", + "size": 913043907, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__Mel_Band_Roformer_Full_Scratch/model_vocals.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__Mel_Band_Roformer_Full_Scratch/model_vocals.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__Mel_Band_Roformer_Full_Scratch/model_vocals.ckpt", + "size": 725806842, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__Mel_Band_Roformer_small/mel_band_roformer_small.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__Mel_Band_Roformer_small/mel_band_roformer_small.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__Mel_Band_Roformer_small/mel_band_roformer_small.ckpt", + "size": 202573672, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__melbandroformer4stems/mel_band_roformer_4stems_large_ver1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__melbandroformer4stems/mel_band_roformer_4stems_large_ver1.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__melbandroformer4stems/mel_band_roformer_4stems_large_ver1.ckpt", + "size": 3764030874, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Aname-Tommy__melbandroformer4stems/mel_band_roformer_4stems_xl_ver1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Aname-Tommy__melbandroformer4stems/mel_band_roformer_4stems_xl_ver1.ckpt", + "relpath": "downloaded/huggingface/Aname-Tommy__melbandroformer4stems/mel_band_roformer_4stems_xl_ver1.ckpt", + "size": 6411087858, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/anvuew__BS-RoFormer/bs_roformer_anvuew_sdr_12.45.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/anvuew__BS-RoFormer/bs_roformer_anvuew_sdr_12.45.ckpt", + "relpath": "downloaded/huggingface/anvuew__BS-RoFormer/bs_roformer_anvuew_sdr_12.45.ckpt", + "size": 204485563, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/anvuew__BS-RoFormer/bs_roformer_ft1_anvuew_sdr_12.55.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/anvuew__BS-RoFormer/bs_roformer_ft1_anvuew_sdr_12.55.ckpt", + "relpath": "downloaded/huggingface/anvuew__BS-RoFormer/bs_roformer_ft1_anvuew_sdr_12.55.ckpt", + "size": 204493312, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/anvuew__BS_RoFormer_mag/bs_roformer_mag_anvuew.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/anvuew__BS_RoFormer_mag/bs_roformer_mag_anvuew.ckpt", + "relpath": "downloaded/huggingface/anvuew__BS_RoFormer_mag/bs_roformer_mag_anvuew.ckpt", + "size": 204502577, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/anvuew__dereverb_bs_roformer/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/anvuew__dereverb_bs_roformer/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt", + "relpath": "downloaded/huggingface/anvuew__dereverb_bs_roformer/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt", + "size": 204496717, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/anvuew__dereverb_mel_band_roformer/archive_only/8_256_6/deverb_mel_band_roformer_8_256dim_6depth.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/anvuew__dereverb_mel_band_roformer/archive_only/8_256_6/deverb_mel_band_roformer_8_256dim_6depth.ckpt", + "relpath": "downloaded/huggingface/anvuew__dereverb_mel_band_roformer/archive_only/8_256_6/deverb_mel_band_roformer_8_256dim_6depth.ckpt", + "size": 438988833, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/anvuew__dereverb_mel_band_roformer/archive_only/8_512_12/deverb_mel_band_roformer_8_512dim_12depth.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/anvuew__dereverb_mel_band_roformer/archive_only/8_512_12/deverb_mel_band_roformer_8_512dim_12depth.ckpt", + "relpath": "downloaded/huggingface/anvuew__dereverb_mel_band_roformer/archive_only/8_512_12/deverb_mel_band_roformer_8_512dim_12depth.ckpt", + "size": 1708522165, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/anvuew__dereverb_room/dereverb_room_anvuew_sdr_13.7432.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/anvuew__dereverb_room/dereverb_room_anvuew_sdr_13.7432.ckpt", + "relpath": "downloaded/huggingface/anvuew__dereverb_room/dereverb_room_anvuew_sdr_13.7432.ckpt", + "size": 118128452, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/armyshope__bs_mamba2/model_bs_mamba2_ep_11_sdr_6.8723.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/armyshope__bs_mamba2/model_bs_mamba2_ep_11_sdr_6.8723.ckpt", + "relpath": "downloaded/huggingface/armyshope__bs_mamba2/model_bs_mamba2_ep_11_sdr_6.8723.ckpt", + "size": 142774366, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/becruily__mel-band-roformer-deux/becruily_deux.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/becruily__mel-band-roformer-deux/becruily_deux.ckpt", + "relpath": "downloaded/huggingface/becruily__mel-band-roformer-deux/becruily_deux.ckpt", + "size": 435006815, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/becruily__mel-band-roformer-guitar/becruily_guitar.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/becruily__mel-band-roformer-guitar/becruily_guitar.ckpt", + "relpath": "downloaded/huggingface/becruily__mel-band-roformer-guitar/becruily_guitar.ckpt", + "size": 45142183, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/ChanTrail__BS-RoFormer/logic_bs_roformer.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/ChanTrail__BS-RoFormer/logic_bs_roformer.ckpt", + "relpath": "downloaded/huggingface/ChanTrail__BS-RoFormer/logic_bs_roformer.ckpt", + "size": 699446752, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/CreoAI__RoFormer/model_melband_roformer_ep_124_si_sdr_increase_7.6621.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/CreoAI__RoFormer/model_melband_roformer_ep_124_si_sdr_increase_7.6621.ckpt", + "relpath": "downloaded/huggingface/CreoAI__RoFormer/model_melband_roformer_ep_124_si_sdr_increase_7.6621.ckpt", + "size": 4645148401, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/DryPaintMan__MelBandRoformer-Duet/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/DryPaintMan__MelBandRoformer-Duet/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt", + "relpath": "downloaded/huggingface/DryPaintMan__MelBandRoformer-Duet/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt", + "size": 630204333, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-BSRoFormer/model_bsrofomer_essid_vocals.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-BSRoFormer/model_bsrofomer_essid_vocals.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-BSRoFormer/model_bsrofomer_essid_vocals.ckpt", + "size": 639330302, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-MDX23C/model_mdx23c_ep_0_sdr_14.6632.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-MDX23C/model_mdx23c_ep_0_sdr_14.6632.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-MDX23C/model_mdx23c_ep_0_sdr_14.6632.ckpt", + "size": 448103606, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-MelBandRoformer/essid_mel_inst.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_inst.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_inst.ckpt", + "size": 913106900, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-MelBandRoformer/essid_mel_inst_old.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_inst_old.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_inst_old.ckpt", + "size": 913107578, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-MelBandRoformer/essid_mel_vocal_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_vocal_v1.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_vocal_v1.ckpt", + "size": 913106900, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-MelBandRoformer/essid_mel_vocal_v2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_vocal_v2.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-MelBandRoformer/essid_mel_vocal_v2.ckpt", + "size": 913106900, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-MelBandRoformer/last_mel_band_roformer.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-MelBandRoformer/last_mel_band_roformer.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-MelBandRoformer/last_mel_band_roformer.ckpt", + "size": 2739327117, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Essid-MelBandRoformer/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Essid-MelBandRoformer/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt", + "relpath": "downloaded/huggingface/Essid__Essid-MelBandRoformer/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt", + "size": 2739362141, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Essid__Mel-Roformer-Inst-FT-Just1Epoch/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Essid__Mel-Roformer-Inst-FT-Just1Epoch/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt", + "relpath": "downloaded/huggingface/Essid__Mel-Roformer-Inst-FT-Just1Epoch/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt", + "size": 913097300, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/bsroformers/bs_karaoke_gabox_IS.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/bsroformers/bs_karaoke_gabox_IS.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/bsroformers/bs_karaoke_gabox_IS.ckpt", + "size": 204483448, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/bsroformers/last_bs_roformer.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/bsroformers/last_bs_roformer.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/bsroformers/last_bs_roformer.ckpt", + "size": 204483448, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Fullness.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Fullness.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Fullness.ckpt", + "size": 913091027, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/inst_fv7b.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/inst_fv7b.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/inst_fv7b.ckpt", + "size": 913030778, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Inst_Fv8.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Inst_Fv8.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Inst_Fv8.ckpt", + "size": 913027055, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Inst_Fv9.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Inst_Fv9.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Inst_Fv9.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV10.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV10.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV10.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7beta2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7beta2.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7beta2.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7beta3.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7beta3.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7beta3.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7plus.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7plus.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/instv7plus.ckpt", + "size": 913031195, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV8.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV8.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV8.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV9.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV9.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/INSTV9.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Lead_VocalDereverb.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Lead_VocalDereverb.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/Lead_VocalDereverb.ckpt", + "size": 913031195, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/small_inst.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/small_inst.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/small_inst.ckpt", + "size": 202573672, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/vocfv7beta2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/vocfv7beta2.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/vocfv7beta2.ckpt", + "size": 913030778, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/vocfv7beta3.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/vocfv7beta3.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/experimental/vocfv7beta3.ckpt", + "size": 913030778, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_ExperimentalV1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_ExperimentalV1.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_ExperimentalV1.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/inst_gaboxFlowersV10.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/inst_gaboxFlowersV10.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/inst_gaboxFlowersV10.ckpt", + "size": 489571079, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_GaboxFv9.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_GaboxFv9.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_GaboxFv9.ckpt", + "size": 913091027, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_GaboxV7.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_GaboxV7.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/instrumental/Inst_GaboxV7.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/karaoke/Karaoke_GaboxV1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/karaoke/Karaoke_GaboxV1.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/karaoke/Karaoke_GaboxV1.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/karaoke/small_karaoke_gaboxaufr.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/karaoke/small_karaoke_gaboxaufr.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/karaoke/small_karaoke_gaboxaufr.ckpt", + "size": 202574227, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/GaboxR67__MelBandRoformers/melbandroformers/vocals/voc_fv7.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/vocals/voc_fv7.ckpt", + "relpath": "downloaded/huggingface/GaboxR67__MelBandRoformers/melbandroformers/vocals/voc_fv7.ckpt", + "size": 489571079, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__BS-RoFormer-BowedStrings-Duality/gilliaan_bowedstrings_bs_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__BS-RoFormer-BowedStrings-Duality/gilliaan_bowedstrings_bs_v1.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__BS-RoFormer-BowedStrings-Duality/gilliaan_bowedstrings_bs_v1.ckpt", + "size": 303490678, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__BS-RoFormer-DrumsOther-Duality/gilliaan_drumsV1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__BS-RoFormer-DrumsOther-Duality/gilliaan_drumsV1.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__BS-RoFormer-DrumsOther-Duality/gilliaan_drumsV1.ckpt", + "size": 303490678, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__BS-RoFormer-Phantom-Center-Wide-Extractor/CenterExtractorSingleStem/gilliaan_CenterExtractor_Single_StemV1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__BS-RoFormer-Phantom-Center-Wide-Extractor/CenterExtractorSingleStem/gilliaan_CenterExtractor_Single_StemV1.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__BS-RoFormer-Phantom-Center-Wide-Extractor/CenterExtractorSingleStem/gilliaan_CenterExtractor_Single_StemV1.ckpt", + "size": 204495806, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__BS-RoFormer-Phantom-Center-Wide-Extractor/CenterExtractorSingleStem/gilliaan_CenterExtractor_Single_StemV2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__BS-RoFormer-Phantom-Center-Wide-Extractor/CenterExtractorSingleStem/gilliaan_CenterExtractor_Single_StemV2.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__BS-RoFormer-Phantom-Center-Wide-Extractor/CenterExtractorSingleStem/gilliaan_CenterExtractor_Single_StemV2.ckpt", + "size": 204435682, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__MDX23C-Phantom-Center-Wide-Extractor/gilliaan_mdx23c_centerwide_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__MDX23C-Phantom-Center-Wide-Extractor/gilliaan_mdx23c_centerwide_v1.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__MDX23C-Phantom-Center-Wide-Extractor/gilliaan_mdx23c_centerwide_v1.ckpt", + "size": 437617081, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__Mel-Band-Roformer-MonoStereo-Duality/gilliaan_MonoStereo_Dual_Beta1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__Mel-Band-Roformer-MonoStereo-Duality/gilliaan_MonoStereo_Dual_Beta1.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__Mel-Band-Roformer-MonoStereo-Duality/gilliaan_MonoStereo_Dual_Beta1.ckpt", + "size": 869726129, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__Mel-Band-Roformer-MonoStereo-Duality/gilliaan_MonoStereo_Dual_Beta2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__Mel-Band-Roformer-MonoStereo-Duality/gilliaan_MonoStereo_Dual_Beta2.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__Mel-Band-Roformer-MonoStereo-Duality/gilliaan_MonoStereo_Dual_Beta2.ckpt", + "size": 869726129, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/gilliaaan__SCNet-Phantom-Center-Wide-Extractor/gilliaan_scnet_centerwide_dual.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/gilliaaan__SCNet-Phantom-Center-Wide-Extractor/gilliaan_scnet_centerwide_dual.ckpt", + "relpath": "downloaded/huggingface/gilliaaan__SCNet-Phantom-Center-Wide-Extractor/gilliaan_scnet_centerwide_dual.ckpt", + "size": 214021674, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Gonzaluigi__Mel-Band-Karaoke-Fusion/mel_band_karaoke_fusion_aggressive.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Gonzaluigi__Mel-Band-Karaoke-Fusion/mel_band_karaoke_fusion_aggressive.ckpt", + "relpath": "downloaded/huggingface/Gonzaluigi__Mel-Band-Karaoke-Fusion/mel_band_karaoke_fusion_aggressive.ckpt", + "size": 913091027, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Gonzaluigi__Mel-Band-Karaoke-Fusion/mel_band_karaoke_fusion_standard.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Gonzaluigi__Mel-Band-Karaoke-Fusion/mel_band_karaoke_fusion_standard.ckpt", + "relpath": "downloaded/huggingface/Gonzaluigi__Mel-Band-Karaoke-Fusion/mel_band_karaoke_fusion_standard.ckpt", + "size": 913091027, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Gonzaluigi__Mel-Band-Roformer-BVE-Gonzaluigi/mel_band_roformer_bve_gonza.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Gonzaluigi__Mel-Band-Roformer-BVE-Gonzaluigi/mel_band_roformer_bve_gonza.ckpt", + "relpath": "downloaded/huggingface/Gonzaluigi__Mel-Band-Roformer-BVE-Gonzaluigi/mel_band_roformer_bve_gonza.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-cmn_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-cmn_state_dict.ckpt", + "relpath": "downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-cmn_state_dict.ckpt", + "size": 149097604, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-deu_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-deu_state_dict.ckpt", + "relpath": "downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-deu_state_dict.ckpt", + "size": 149097604, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-eng_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-eng_state_dict.ckpt", + "relpath": "downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-eng_state_dict.ckpt", + "size": 149097604, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-fao_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-fao_state_dict.ckpt", + "relpath": "downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-fao_state_dict.ckpt", + "size": 149097604, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-fra_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-fra_state_dict.ckpt", + "relpath": "downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-fra_state_dict.ckpt", + "size": 149097604, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-multi_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-multi_state_dict.ckpt", + "relpath": "downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-multi_state_dict.ckpt", + "size": 149100976, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-spa_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-spa_state_dict.ckpt", + "relpath": "downloaded/huggingface/jarredou__banditv2_state_dicts_only/checkpoint-spa_state_dict.ckpt", + "size": 149097604, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__lew_apollo_vocal_enhancer/apollo_model.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__lew_apollo_vocal_enhancer/apollo_model.ckpt", + "relpath": "downloaded/huggingface/jarredou__lew_apollo_vocal_enhancer/apollo_model.ckpt", + "size": 66532080, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__lew_apollo_vocal_enhancer/apollo_model_v2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__lew_apollo_vocal_enhancer/apollo_model_v2.ckpt", + "relpath": "downloaded/huggingface/jarredou__lew_apollo_vocal_enhancer/apollo_model_v2.ckpt", + "size": 38159792, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jarredou__unwa_bs_roformer/BS-Roformer_LargeV1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jarredou__unwa_bs_roformer/BS-Roformer_LargeV1.ckpt", + "relpath": "downloaded/huggingface/jarredou__unwa_bs_roformer/BS-Roformer_LargeV1.ckpt", + "size": 740263433, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet.ckpt", + "relpath": "downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet.ckpt", + "size": 42458509, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_bass_drums.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_bass_drums.ckpt", + "relpath": "downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_bass_drums.ckpt", + "size": 42458509, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_large.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_large.ckpt", + "relpath": "downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_large.ckpt", + "size": 168869389, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_piano_drums.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_piano_drums.ckpt", + "relpath": "downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_piano_drums.ckpt", + "size": 42458509, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_xl.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_xl.ckpt", + "relpath": "downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_scnet_xl.ckpt", + "size": 216190397, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/jorisvaneyghen__SCNet/model_jazz_split_scnet.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_split_scnet.ckpt", + "relpath": "downloaded/huggingface/jorisvaneyghen__SCNet/model_jazz_split_scnet.ckpt", + "size": 42434765, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/04573f0d-f3cf25b2.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/04573f0d-f3cf25b2.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/04573f0d-f3cf25b2.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/0d19c1c6-0f06f20e.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/0d19c1c6-0f06f20e.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/0d19c1c6-0f06f20e.th", + "size": 178048329, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/14fc6a69-a89dd0ee.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/14fc6a69-a89dd0ee.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/14fc6a69-a89dd0ee.th", + "size": 38491885, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/1ef250f1-592467ce.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/1ef250f1-592467ce.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/1ef250f1-592467ce.th", + "size": 267602121, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/305bc58f-18378783.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/305bc58f-18378783.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/305bc58f-18378783.th", + "size": 46847123, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/42e558d4-196e0e1b.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/42e558d4-196e0e1b.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/42e558d4-196e0e1b.th", + "size": 58227087, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/464b36d7-e5a9386e.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/464b36d7-e5a9386e.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/464b36d7-e5a9386e.th", + "size": 38893153, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/5c90dfd2-34c22ccb.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/5c90dfd2-34c22ccb.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/5c90dfd2-34c22ccb.th", + "size": 54996327, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/5d2d6c55-db83574e.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/5d2d6c55-db83574e.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/5d2d6c55-db83574e.th", + "size": 167391595, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/6b9c2ca1-3fd82607.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/6b9c2ca1-3fd82607.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/6b9c2ca1-3fd82607.th", + "size": 59648321, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/75fc33f5-1941ce65.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/75fc33f5-1941ce65.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/75fc33f5-1941ce65.th", + "size": 167407275, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/7d865c68-3d5dd56b.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/7d865c68-3d5dd56b.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/7d865c68-3d5dd56b.th", + "size": 167918783, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/7ecf8ec1-70f50cc9.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/7ecf8ec1-70f50cc9.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/7ecf8ec1-70f50cc9.th", + "size": 178048329, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/7fd6ef75-a905dd85.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/7fd6ef75-a905dd85.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/7fd6ef75-a905dd85.th", + "size": 39436529, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/83fc094f-4a16d450.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/83fc094f-4a16d450.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/83fc094f-4a16d450.th", + "size": 50756993, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/902315c2-b39ce9c9.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/902315c2-b39ce9c9.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/902315c2-b39ce9c9.th", + "size": 167405611, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/92cfc3b6-ef3bcb9c.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/92cfc3b6-ef3bcb9c.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/92cfc3b6-ef3bcb9c.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/955717e8-8726e21a.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/955717e8-8726e21a.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/955717e8-8726e21a.th", + "size": 84141911, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/9a6b4851-03af0aa6.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/9a6b4851-03af0aa6.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/9a6b4851-03af0aa6.th", + "size": 267602121, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/a1d90b5c-ae9d2452.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/a1d90b5c-ae9d2452.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/a1d90b5c-ae9d2452.th", + "size": 167391595, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/b72baf4e-8778635e.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/b72baf4e-8778635e.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/b72baf4e-8778635e.th", + "size": 44368175, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/BS-Roformer-SW.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/BS-Roformer-SW.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/BS-Roformer-SW.ckpt", + "size": 699412152, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/bs_roformer_instrumental_resurrection_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_instrumental_resurrection_gabox.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_instrumental_resurrection_gabox.ckpt", + "size": 204483033, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/bs_roformer_karaoke_anvuew.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_karaoke_anvuew.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_karaoke_anvuew.ckpt", + "size": 204486925, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/bs_roformer_karaoke_frazer_becruily.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_karaoke_frazer_becruily.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_karaoke_frazer_becruily.ckpt", + "size": 204436907, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt", + "size": 527119779, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/bs_roformer_vocals_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_vocals_gabox.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/bs_roformer_vocals_gabox.ckpt", + "size": 639254584, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/c511e2ab-fe698775.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/c511e2ab-fe698775.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/c511e2ab-fe698775.th", + "size": 167334095, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/cfa93e08-61801ae1.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/cfa93e08-61801ae1.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/cfa93e08-61801ae1.th", + "size": 167399275, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/checkpoint-multi_fixed.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/checkpoint-multi_fixed.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/checkpoint-multi_fixed.ckpt", + "size": 149133378, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/d12395a8-e57c48e6.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/d12395a8-e57c48e6.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/d12395a8-e57c48e6.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/demucs-e07c671f.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/demucs-e07c671f.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/demucs-e07c671f.th", + "size": 1062738817, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/demucs.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/demucs.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/demucs.th", + "size": 2594202563, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/demucs48_hq-28a1282c.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/demucs48_hq-28a1282c.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/demucs48_hq-28a1282c.th", + "size": 597870849, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/demucs_extra-3646af93.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/demucs_extra-3646af93.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/demucs_extra-3646af93.th", + "size": 1062738817, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/demucs_extra.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/demucs_extra.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/demucs_extra.th", + "size": 2594202563, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/demucs_unittest-09ebc15f.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/demucs_unittest-09ebc15f.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/demucs_unittest-09ebc15f.th", + "size": 4194049, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/dereverb_mdx23c_sdr_6.9096.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/dereverb_mdx23c_sdr_6.9096.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/dereverb_mdx23c_sdr_6.9096.ckpt", + "size": 448098867, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/deverb_bs_roformer_8_256dim_8depth.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/deverb_bs_roformer_8_256dim_8depth.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/deverb_bs_roformer_8_256dim_8depth.ckpt", + "size": 170770820, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/e51eebcc-c1b80bdd.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/e51eebcc-c1b80bdd.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/e51eebcc-c1b80bdd.th", + "size": 167399275, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/ebf34a2db.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/ebf34a2db.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/ebf34a2db.th", + "size": 167390507, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/f7e0c4bc-ba3fe64a.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/f7e0c4bc-ba3fe64a.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/f7e0c4bc-ba3fe64a.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/fa0cb7f9-100d8bf4.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/fa0cb7f9-100d8bf4.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/fa0cb7f9-100d8bf4.th", + "size": 167405675, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/light.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/light.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/light.th", + "size": 1062727712, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/light_extra.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/light_extra.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/light_extra.th", + "size": 1062727712, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/MDX23C-8KFFT-InstVoc_HQ_2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/MDX23C-8KFFT-InstVoc_HQ_2.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/MDX23C-8KFFT-InstVoc_HQ_2.ckpt", + "size": 448093770, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/MDX23C_D1581.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/MDX23C_D1581.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/MDX23C_D1581.ckpt", + "size": 183379219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt", + "size": 913091027, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/mel_band_roformer_instrumental_fv8b_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/mel_band_roformer_instrumental_fv8b_gabox.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/mel_band_roformer_instrumental_fv8b_gabox.ckpt", + "size": 913027055, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/mel_band_roformer_vocals_fv7b_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/mel_band_roformer_vocals_fv7b_gabox.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/mel_band_roformer_vocals_fv7b_gabox.ckpt", + "size": 913091027, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/MelBandRoformerBigSYHFTV1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerBigSYHFTV1.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerBigSYHFTV1.ckpt", + "size": 1479738496, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFT.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFT.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFT.ckpt", + "size": 913096702, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV2.5.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV2.5.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV2.5.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV2.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV2.ckpt", + "size": 913095346, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV3Epsilon.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV3Epsilon.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/MelBandRoformerSYHFTV3Epsilon.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/model_bandit_plus_dnr_sdr_11.47.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/model_bandit_plus_dnr_sdr_11.47.ckpt", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/model_bandit_plus_dnr_sdr_11.47.ckpt", + "size": 148891175, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/tasnet-beb46fac.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/tasnet-beb46fac.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/tasnet-beb46fac.th", + "size": 44035981, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/tasnet.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/tasnet.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/tasnet.th", + "size": 43973949, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/tasnet_extra-df3777b2.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/tasnet_extra-df3777b2.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/tasnet_extra-df3777b2.th", + "size": 44035981, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__audio-separator-models/tasnet_extra.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__audio-separator-models/tasnet_extra.th", + "relpath": "downloaded/huggingface/lainlives__audio-separator-models/tasnet_extra.th", + "size": 43973949, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__MelBand_BS_Roformers-pytorch/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__MelBand_BS_Roformers-pytorch/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa.ckpt", + "relpath": "downloaded/huggingface/lainlives__MelBand_BS_Roformers-pytorch/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa/MelBand_Roformer_Kim_InstVoc_Duality_V1_by_Unwa.ckpt", + "size": 299892736, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__MelBand_BS_Roformers-pytorch/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__MelBand_BS_Roformers-pytorch/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa.ckpt", + "relpath": "downloaded/huggingface/lainlives__MelBand_BS_Roformers-pytorch/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa/MelBand_Roformer_Kim_InstVoc_Duality_V2_by_Unwa.ckpt", + "size": 3407872, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__MelBand_BS_Roformers/MelBand_Roformer_Aspiration_by_Sucial/MelBand_Roformer_Aspiration_by_Sucial.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__MelBand_BS_Roformers/MelBand_Roformer_Aspiration_by_Sucial/MelBand_Roformer_Aspiration_by_Sucial.ckpt", + "relpath": "downloaded/huggingface/lainlives__MelBand_BS_Roformers/MelBand_Roformer_Aspiration_by_Sucial/MelBand_Roformer_Aspiration_by_Sucial.ckpt", + "size": 835983746, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__MelBand_BS_Roformers/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__MelBand_BS_Roformers/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial.ckpt", + "relpath": "downloaded/huggingface/lainlives__MelBand_BS_Roformers/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial/MelBand_Roformer_Aspiration_Less_Aggressive_by_Sucial.ckpt", + "size": 835982664, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__MelBand_BS_Roformers/mini-bs-roformer-v2-46.8M/model.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__MelBand_BS_Roformers/mini-bs-roformer-v2-46.8M/model.ckpt", + "relpath": "downloaded/huggingface/lainlives__MelBand_BS_Roformers/mini-bs-roformer-v2-46.8M/model.ckpt", + "size": 94204810, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/lainlives__MelBand_BS_Roformers/mvsep_mega_model/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/lainlives__MelBand_BS_Roformers/mvsep_mega_model/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt", + "relpath": "downloaded/huggingface/lainlives__MelBand_BS_Roformers/mvsep_mega_model/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt", + "size": 1368919887, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/meskvlla33__metal_roformer_preview/metal_roformer_inst_mesk_preview.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/meskvlla33__metal_roformer_preview/metal_roformer_inst_mesk_preview.ckpt", + "relpath": "downloaded/huggingface/meskvlla33__metal_roformer_preview/metal_roformer_inst_mesk_preview.ckpt", + "size": 944683368, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/MrDense267__BS_Roformer_EFENEF/BS_Roformer_FNF_V1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/MrDense267__BS_Roformer_EFENEF/BS_Roformer_FNF_V1.ckpt", + "relpath": "downloaded/huggingface/MrDense267__BS_Roformer_EFENEF/BS_Roformer_FNF_V1.ckpt", + "size": 613995746, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/natanworkspace__bs_roformer/instrumental/neo_inst_beta.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/natanworkspace__bs_roformer/instrumental/neo_inst_beta.ckpt", + "relpath": "downloaded/huggingface/natanworkspace__bs_roformer/instrumental/neo_inst_beta.ckpt", + "size": 204483448, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/natanworkspace__melband_roformer/Neo_InstVFX.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/natanworkspace__melband_roformer/Neo_InstVFX.ckpt", + "relpath": "downloaded/huggingface/natanworkspace__melband_roformer/Neo_InstVFX.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt", + "size": 913097300, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt", + "size": 913097300, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt", + "size": 835997896, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt", + "size": 455862568, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_big_mbr_ep_362.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_big_mbr_ep_362.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_big_mbr_ep_362.ckpt", + "size": 455864012, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_echo_mbr_fused.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_echo_mbr_fused.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_echo_mbr_fused.ckpt", + "size": 455776577, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt", + "size": 913107578, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt", + "size": 913107578, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_mono_anvuew.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_mono_anvuew.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_mel_band_roformer_mono_anvuew.ckpt", + "size": 913097978, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_super_big_mbr_ep_346.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_super_big_mbr_ep_346.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/dereverb_super_big_mbr_ep_346.ckpt", + "size": 455864012, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/deverb_bs_roformer_8_384dim_10depth.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/deverb_bs_roformer_8_384dim_10depth.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/deverb_bs_roformer_8_384dim_10depth.ckpt", + "size": 361499604, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_bleed_suppressor_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_bleed_suppressor_v1.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_bleed_suppressor_v1.ckpt", + "size": 913102724, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt", + "size": 913096801, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_denoise_debleed_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_denoise_debleed_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_denoise_debleed_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_2_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_2_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_2_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_3_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_3_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_3_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_becruily.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_becruily.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_becruily.ckpt", + "size": 913106900, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fv7z_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fv7z_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fv7z_gabox.ckpt", + "size": 913031195, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fv8_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fv8_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fv8_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fvx_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fvx_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_fvx_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv5_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv5_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv5_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv5n_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv5n_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv5n_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv6_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv6_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv6_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv6n_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv6n_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv6n_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv7_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv7_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv7_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv7n_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv7n_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv7n_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv8_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv8_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv8_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv8n_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv8n_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_instrumental_instv8n_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_becruily.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_becruily.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_becruily.ckpt", + "size": 1719139254, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_gabox_v2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_gabox_v2.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_karaoke_gabox_v2.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocal_fullness_aname.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocal_fullness_aname.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocal_fullness_aname.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_becruily.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_becruily.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_becruily.ckpt", + "size": 913107578, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv1_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv1_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv1_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv2_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv2_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv2_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv3_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv3_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv3_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv4_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv4_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv4_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv5_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv5_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv5_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv6_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv6_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_fv6_gabox.ckpt", + "size": 913031195, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_gabox.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_v2_gabox.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_v2_gabox.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/mel_band_roformer_vocals_v2_gabox.ckpt", + "size": 913031195, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_317_sdr_12.9755.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_317_sdr_12.9755.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_317_sdr_12.9755.ckpt", + "size": 639331213, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_368_sdr_12.9628.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_368_sdr_12.9628.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_368_sdr_12.9628.ckpt", + "size": 639317465, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_937_sdr_10.5309.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_937_sdr_10.5309.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_bs_roformer_ep_937_sdr_10.5309.ckpt", + "size": 393068365, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt", + "size": 527121477, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/audio-separator-models/roformers/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt", + "size": 1007816988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/mel_band_roformers/mel_band_roformer_karaoke.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/mel_band_roformers/mel_band_roformer_karaoke.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/mel_band_roformers/mel_band_roformer_karaoke.ckpt", + "size": 913096801, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/niobures__RoFormer/models/mel_band_roformers/mel_band_roformer_vocals.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/niobures__RoFormer/models/mel_band_roformers/mel_band_roformer_vocals.ckpt", + "relpath": "downloaded/huggingface/niobures__RoFormer/models/mel_band_roformers/mel_band_roformer_vocals.ckpt", + "size": 913106900, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__all_models_for_mel_band_roformer/gabox_karaoke_25_02_2025.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__all_models_for_mel_band_roformer/gabox_karaoke_25_02_2025.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__all_models_for_mel_band_roformer/gabox_karaoke_25_02_2025.ckpt", + "size": 913026650, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_accordion_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_accordion_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_accordion_mvsep.ckpt", + "size": 77619808, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_acoustic-guitar_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_acoustic-guitar_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_acoustic-guitar_mvsep.ckpt", + "size": 77624038, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_back-vocal_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_back-vocal_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_back-vocal_mvsep.ckpt", + "size": 77620513, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_banjo_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_banjo_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_banjo_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bass_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bass_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bass_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bassoon_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bassoon_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bassoon_mvsep.ckpt", + "size": 77618398, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bells_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bells_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bells_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bowed_strings_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bowed_strings_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_bowed_strings_mvsep.ckpt", + "size": 77622628, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_brass_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_brass_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_brass_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_cello_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_cello_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_cello_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_clarinet_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_clarinet_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_clarinet_mvsep.ckpt", + "size": 77619103, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_congas_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_congas_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_congas_mvsep.ckpt", + "size": 77617693, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_digital-piano_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_digital-piano_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_digital-piano_mvsep.ckpt", + "size": 77622628, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_dobro_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_dobro_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_dobro_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_double-bass_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_double-bass_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_double-bass_mvsep.ckpt", + "size": 77621218, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_drums_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_drums_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_drums_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_electric-guitar_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_electric-guitar_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_electric-guitar_mvsep.ckpt", + "size": 77624038, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_flute_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_flute_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_flute_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_french-horn_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_french-horn_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_french-horn_mvsep.ckpt", + "size": 77621218, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_glockenspiel_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_glockenspiel_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_glockenspiel_mvsep.ckpt", + "size": 77621923, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_guitar_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_guitar_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_guitar_mvsep.ckpt", + "size": 77617693, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harmonica_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harmonica_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harmonica_mvsep.ckpt", + "size": 77619808, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harp_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harp_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harp_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harpsichord_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harpsichord_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_harpsichord_mvsep.ckpt", + "size": 77621218, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_hh_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_hh_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_hh_mvsep.ckpt", + "size": 77614809, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_keys_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_keys_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_keys_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_kick_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_kick_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_kick_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_lead-vocal_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_lead-vocal_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_lead-vocal_mvsep.ckpt", + "size": 77620513, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_mandolin_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_mandolin_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_mandolin_mvsep.ckpt", + "size": 77619103, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_marimba_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_marimba_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_marimba_mvsep.ckpt", + "size": 77618398, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_oboe_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_oboe_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_oboe_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_organ_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_organ_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_organ_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_percussion_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_percussion_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_percussion_mvsep.ckpt", + "size": 77620513, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_piano_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_piano_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_piano_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_saxophone_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_saxophone_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_saxophone_mvsep.ckpt", + "size": 77619808, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_sitar_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_sitar_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_sitar_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_snare_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_snare_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_snare_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_strings_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_strings_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_strings_mvsep.ckpt", + "size": 77618398, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_synth_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_synth_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_synth_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_tambourine_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_tambourine_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_tambourine_mvsep.ckpt", + "size": 77620513, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_timpani_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_timpani_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_timpani_mvsep.ckpt", + "size": 77618398, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_toms_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_toms_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_toms_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_triangle_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_triangle_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_triangle_mvsep.ckpt", + "size": 77619103, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_trombone_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_trombone_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_trombone_mvsep.ckpt", + "size": 77619103, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_trumpet_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_trumpet_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_trumpet_mvsep.ckpt", + "size": 77618398, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_tuba_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_tuba_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_tuba_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_ukulele_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_ukulele_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_ukulele_mvsep.ckpt", + "size": 77618398, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_viola_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_viola_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_viola_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_violin_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_violin_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_violin_mvsep.ckpt", + "size": 77617693, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_vocal_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_vocal_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_vocal_mvsep.ckpt", + "size": 77616988, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_wind-chimes_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_wind-chimes_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_wind-chimes_mvsep.ckpt", + "size": 77621218, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_wind_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_wind_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_wind_mvsep.ckpt", + "size": 77616219, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_woodwind_mvsep.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_woodwind_mvsep.ckpt", + "relpath": "downloaded/huggingface/noblebarkrr__BS-Roformer-MVSep-Mega-53-stems/v1/bs_mega_53stem_woodwind_mvsep.ckpt", + "size": 77619103, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/oulianov__Apollo-vocal-msst/epoch=54-val_loss=-17.6221.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/oulianov__Apollo-vocal-msst/epoch=54-val_loss=-17.6221.ckpt", + "relpath": "downloaded/huggingface/oulianov__Apollo-vocal-msst/epoch=54-val_loss=-17.6221.ckpt", + "size": 615867827, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/oulianov__Apollo-vocal-msst/model_apollo_vocals_ep_54.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/oulianov__Apollo-vocal-msst/model_apollo_vocals_ep_54.ckpt", + "relpath": "downloaded/huggingface/oulianov__Apollo-vocal-msst/model_apollo_vocals_ep_54.ckpt", + "size": 194278526, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-EXP-SiameseRoformer/bs_exp_siameseroformer.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-EXP-SiameseRoformer/bs_exp_siameseroformer.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-EXP-SiameseRoformer/bs_exp_siameseroformer.ckpt", + "size": 204682749, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-HyperACE/bs_hyperace.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-HyperACE/bs_hyperace.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-HyperACE/bs_hyperace.ckpt", + "size": 274890855, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-HyperACE/v2_inst/bs_roformer_inst_hyperacev2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-HyperACE/v2_inst/bs_roformer_inst_hyperacev2.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-HyperACE/v2_inst/bs_roformer_inst_hyperacev2.ckpt", + "size": 288725646, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-HyperACE/v2_voc/bs_roformer_voc_hyperacev2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-HyperACE/v2_voc/bs_roformer_voc_hyperacev2.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-HyperACE/v2_voc/bs_roformer_voc_hyperacev2.ckpt", + "size": 288724430, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Inst-EXP-Value-Residual/BS_Inst_EXP_VRL.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Inst-EXP-Value-Residual/BS_Inst_EXP_VRL.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Inst-EXP-Value-Residual/BS_Inst_EXP_VRL.ckpt", + "size": 393351501, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Inst-FNO/bs_roformer_fno.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Inst-FNO/bs_roformer_fno.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Inst-FNO/bs_roformer_fno.ckpt", + "size": 332004435, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Large-Inst/bs_large_v2_inst.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Large-Inst/bs_large_v2_inst.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Large-Inst/bs_large_v2_inst.ckpt", + "size": 238214371, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Resurrection/BS-Roformer-Resurrection-Inst.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Resurrection/BS-Roformer-Resurrection-Inst.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Resurrection/BS-Roformer-Resurrection-Inst.ckpt", + "size": 204483033, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Resurrection/BS-Roformer-Resurrection.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Resurrection/BS-Roformer-Resurrection.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Resurrection/BS-Roformer-Resurrection.ckpt", + "size": 204510749, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive.ckpt", + "size": 639326600, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive2.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive2.ckpt", + "size": 639326600, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive3e.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive3e.ckpt", + "relpath": "downloaded/huggingface/pcunwa__BS-Roformer-Revive/bs_roformer_revive3e.ckpt", + "size": 639326600, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft.ckpt", + "size": 913100690, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft2.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft2.ckpt", + "size": 913100690, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft2_bleedless.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft2_bleedless.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft2_bleedless.ckpt", + "size": 913101368, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft3_prev.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft3_prev.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Kim-Mel-Band-RoFormer-FT/kimmel_unwa_ft3_prev.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta5e.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta5e.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta5e.ckpt", + "size": 1479749810, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta6.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta6.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta6.ckpt", + "size": 1557078584, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta6x.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta6x.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta6x.ckpt", + "size": 1708527586, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta7.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta7.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/big_beta7.ckpt", + "size": 944675923, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta1.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta1.ckpt", + "size": 1479745022, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta2.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta2.ckpt", + "size": 1479745022, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta3.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta3.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta3.ckpt", + "size": 1479738496, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta4.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta4.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-big/melband_roformer_big_beta4.ckpt", + "size": 1574477088, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1_plus_test.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1_plus_test.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1_plus_test.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1e.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1e.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1e.ckpt", + "size": 913102724, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1e_plus.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1e_plus.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/inst_v1e_plus.ckpt", + "size": 913090472, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-Inst/melband_roformer_inst_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/melband_roformer_inst_v1.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/melband_roformer_inst_v1.ckpt", + "size": 913100690, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-Inst/melband_roformer_inst_v2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/melband_roformer_inst_v2.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-Inst/melband_roformer_inst_v2.ckpt", + "size": 1574477088, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-InstVoc-Duality/melband_roformer_instvoc_duality_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-InstVoc-Duality/melband_roformer_instvoc_duality_v1.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-InstVoc-Duality/melband_roformer_instvoc_duality_v1.ckpt", + "size": 1719116358, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-InstVoc-Duality/melband_roformer_instvox_duality_v2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-InstVoc-Duality/melband_roformer_instvox_duality_v2.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-InstVoc-Duality/melband_roformer_instvox_duality_v2.ckpt", + "size": 1719116358, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/pcunwa__Mel-Band-RoFormer-small/melband_roformer_small_v1.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/pcunwa__Mel-Band-RoFormer-small/melband_roformer_small_v1.ckpt", + "relpath": "downloaded/huggingface/pcunwa__Mel-Band-RoFormer-small/melband_roformer_small_v1.ckpt", + "size": 202583890, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/rbustos89__scnetchoir/model_scnet_ep_36_sdr_5.4596.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/rbustos89__scnetchoir/model_scnet_ep_36_sdr_5.4596.ckpt", + "relpath": "downloaded/huggingface/rbustos89__scnetchoir/model_scnet_ep_36_sdr_5.4596.ckpt", + "size": 42457314, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/rbustos89__scnetchoir/model_scnet_ep_42_sdr_5.2559.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/rbustos89__scnetchoir/model_scnet_ep_42_sdr_5.2559.ckpt", + "relpath": "downloaded/huggingface/rbustos89__scnetchoir/model_scnet_ep_42_sdr_5.2559.ckpt", + "size": 42457314, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/rbustos89__scnetmskedchoir/model_scnet_masked_ep_14_sdr_6.3693.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/rbustos89__scnetmskedchoir/model_scnet_masked_ep_14_sdr_6.3693.ckpt", + "relpath": "downloaded/huggingface/rbustos89__scnetmskedchoir/model_scnet_masked_ep_14_sdr_6.3693.ckpt", + "size": 42522921, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/rbustos89__scnetmskedchoir/model_scnet_masked_ep_14_sdr_6.9052.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/rbustos89__scnetmskedchoir/model_scnet_masked_ep_14_sdr_6.9052.ckpt", + "relpath": "downloaded/huggingface/rbustos89__scnetmskedchoir/model_scnet_masked_ep_14_sdr_6.9052.ckpt", + "size": 42522921, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/smulelabs__windowed-roformer/mbr-win10-sink8.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/smulelabs__windowed-roformer/mbr-win10-sink8.ckpt", + "relpath": "downloaded/huggingface/smulelabs__windowed-roformer/mbr-win10-sink8.ckpt", + "size": 913837877, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__Chorus_Male_Female_BS_Roformer/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__Chorus_Male_Female_BS_Roformer/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt", + "relpath": "downloaded/huggingface/Sucial__Chorus_Male_Female_BS_Roformer/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt", + "size": 527121477, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/bs_roformer_4stems_ft.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/bs_roformer_4stems_ft.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/bs_roformer_4stems_ft.ckpt", + "size": 527245586, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/HTDemucs4.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/HTDemucs4.th", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/HTDemucs4.th", + "size": 84141911, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/HTDemucs4_6stems.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/HTDemucs4_6stems.th", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/multi_stem_models/HTDemucs4_6stems.th", + "size": 54996327, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/Apollo_LQ_MP3_restoration.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/Apollo_LQ_MP3_restoration.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/Apollo_LQ_MP3_restoration.ckpt", + "size": 66541845, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/apollo_model_uni.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/apollo_model_uni.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/apollo_model_uni.ckpt", + "size": 146867952, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt", + "size": 835982664, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/aspiration_mel_band_roformer_sdr_18.9845.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/aspiration_mel_band_roformer_sdr_18.9845.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/aspiration_mel_band_roformer_sdr_18.9845.ckpt", + "size": 835983746, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/de_big_reverb_mbr_ep_362.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/de_big_reverb_mbr_ep_362.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/de_big_reverb_mbr_ep_362.ckpt", + "size": 455864012, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt", + "size": 455776577, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt", + "size": 489578757, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_bass.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_bass.th", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_bass.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_drums.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_drums.th", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_drums.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_other.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_other.th", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_other.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_vocals_official.th", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_vocals_official.th", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/single_stem_models/HTDemucs4_FT_vocals_official.th", + "size": 84141271, + "suffix": ".th" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/vocal_models/Kim_MelBandRoformer.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/vocal_models/Kim_MelBandRoformer.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/vocal_models/Kim_MelBandRoformer.ckpt", + "size": 913106900, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/Sucial__MSST-WebUI/All_Models/vocal_models/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/Sucial__MSST-WebUI/All_Models/vocal_models/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt", + "relpath": "downloaded/huggingface/Sucial__MSST-WebUI/All_Models/vocal_models/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt", + "size": 913096801, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/SYH99999__MelBandRoformer4StemFTLarge/MelBandRoformer4StemFTLarge.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/SYH99999__MelBandRoformer4StemFTLarge/MelBandRoformer4StemFTLarge.ckpt", + "relpath": "downloaded/huggingface/SYH99999__MelBandRoformer4StemFTLarge/MelBandRoformer4StemFTLarge.ckpt", + "size": 3764030874, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/SYH99999__MelBandRoformer4StemFTLarge/ver2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/SYH99999__MelBandRoformer4StemFTLarge/ver2.ckpt", + "relpath": "downloaded/huggingface/SYH99999__MelBandRoformer4StemFTLarge/ver2.ckpt", + "size": 3764030874, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/SYH99999__MelBandRoformerMergedSYHFTBeta1/merge_syhft.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/SYH99999__MelBandRoformerMergedSYHFTBeta1/merge_syhft.ckpt", + "relpath": "downloaded/huggingface/SYH99999__MelBandRoformerMergedSYHFTBeta1/merge_syhft.ckpt", + "size": 913036186, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/SYH99999__MelBandRoformerSYHFTB1/model.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/SYH99999__MelBandRoformerSYHFTB1/model.ckpt", + "relpath": "downloaded/huggingface/SYH99999__MelBandRoformerSYHFTB1/model.ckpt", + "size": 913106158, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/SYH99999__MelBandRoformerSYHFTB1/model2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/SYH99999__MelBandRoformerSYHFTB1/model2.ckpt", + "relpath": "downloaded/huggingface/SYH99999__MelBandRoformerSYHFTB1/model2.ckpt", + "size": 913106158, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/SYH99999__MelBandRoformerSYHFTB1/model3.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/SYH99999__MelBandRoformerSYHFTB1/model3.ckpt", + "relpath": "downloaded/huggingface/SYH99999__MelBandRoformerSYHFTB1/model3.ckpt", + "size": 913106158, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/yongyizang__16k_Vocal_Lightweight_MelBandRoformer/model.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/yongyizang__16k_Vocal_Lightweight_MelBandRoformer/model.ckpt", + "relpath": "downloaded/huggingface/yongyizang__16k_Vocal_Lightweight_MelBandRoformer/model.ckpt", + "size": 59960415, + "suffix": ".ckpt" + }, + { + "source_name": "downloaded", + "source": "/Volumes/2T/pymss/benchmark_results/model_downloads/huggingface/yuluoye__model_mel_band_roformer_ep_3_sdr_14.8359/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt", + "dest": "/Volumes/2T/pymss/all_models/downloaded/huggingface/yuluoye__model_mel_band_roformer_ep_3_sdr_14.8359/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt", + "relpath": "downloaded/huggingface/yuluoye__model_mel_band_roformer_ep_3_sdr_14.8359/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt", + "size": 2739398173, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/BS-Roformer-HyperACE_v2_voc/bs_roformer_voc_hyperacev2.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/BS-Roformer-HyperACE_v2_voc/bs_roformer_voc_hyperacev2.ckpt", + "relpath": "manual_models/BS-Roformer-HyperACE_v2_voc/bs_roformer_voc_hyperacev2.ckpt", + "size": 288724430, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/bs_roformer_ep_368/model_bs_roformer_ep_368_sdr_12.9628.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/bs_roformer_ep_368/model_bs_roformer_ep_368_sdr_12.9628.ckpt", + "relpath": "manual_models/bs_roformer_ep_368/model_bs_roformer_ep_368_sdr_12.9628.ckpt", + "size": 639317465, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/logic_bs_roformer/logic_bs_roformer.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/logic_bs_roformer/logic_bs_roformer.ckpt", + "relpath": "manual_models/logic_bs_roformer/logic_bs_roformer.ckpt", + "size": 699446752, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/Mel-Band-Roformer-big/big_beta7.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/Mel-Band-Roformer-big/big_beta7.ckpt", + "relpath": "manual_models/Mel-Band-Roformer-big/big_beta7.ckpt", + "size": 944675923, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/mel-band-roformer-deux/becruily_deux.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/mel-band-roformer-deux/becruily_deux.ckpt", + "relpath": "manual_models/mel-band-roformer-deux/becruily_deux.ckpt", + "size": 435006815, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/model_htdemucs_ep_21_sdr_13.6970.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/model_htdemucs_ep_21_sdr_13.6970.ckpt", + "relpath": "manual_models/model_htdemucs_ep_21_sdr_13.6970.ckpt", + "size": 168122809, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/model_scnet_ep_102_sdr_12.8941.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/model_scnet_ep_102_sdr_12.8941.ckpt", + "relpath": "manual_models/model_scnet_ep_102_sdr_12.8941.ckpt", + "size": 42422546, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/model_scnet_ep_30_sdr_15.1291.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/model_scnet_ep_30_sdr_15.1291.ckpt", + "relpath": "manual_models/model_scnet_ep_30_sdr_15.1291.ckpt", + "size": 42422170, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/apollo/Apollo_LQ_MP3_restoration.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/apollo/Apollo_LQ_MP3_restoration.ckpt", + "relpath": "manual_models/smoke/apollo/Apollo_LQ_MP3_restoration.ckpt", + "size": 66541845, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/apollo/Apollo_LQ_MP3_restoration.pymss_state_dict.pt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/apollo/Apollo_LQ_MP3_restoration.pymss_state_dict.pt", + "relpath": "manual_models/smoke/apollo/Apollo_LQ_MP3_restoration.pymss_state_dict.pt", + "size": 66528927, + "suffix": ".pt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/bandit/model_bandit_plus_dnr_sdr_11.47.chpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/bandit/model_bandit_plus_dnr_sdr_11.47.chpt", + "relpath": "manual_models/smoke/bandit/model_bandit_plus_dnr_sdr_11.47.chpt", + "size": 148891175, + "suffix": ".chpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/bandit_v2/checkpoint-multi_state_dict.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/bandit_v2/checkpoint-multi_state_dict.ckpt", + "relpath": "manual_models/smoke/bandit_v2/checkpoint-multi_state_dict.ckpt", + "size": 149100976, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/htdemucs/HTDemucs4.th", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/htdemucs/HTDemucs4.th", + "relpath": "manual_models/smoke/htdemucs/HTDemucs4.th", + "size": 84141911, + "suffix": ".th" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/mdx23c/model_vocals_mdx23c_sdr_10.17.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/mdx23c/model_vocals_mdx23c_sdr_10.17.ckpt", + "relpath": "manual_models/smoke/mdx23c/model_vocals_mdx23c_sdr_10.17.ckpt", + "size": 448101203, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/scnet/scnet_checkpoint_musdb18.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/scnet/scnet_checkpoint_musdb18.ckpt", + "relpath": "manual_models/smoke/scnet/scnet_checkpoint_musdb18.ckpt", + "size": 42434986, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/segm_models/model_vocals_segm_models_sdr_9.77.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/segm_models/model_vocals_segm_models_sdr_9.77.ckpt", + "relpath": "manual_models/smoke/segm_models/model_vocals_segm_models_sdr_9.77.ckpt", + "size": 863683537, + "suffix": ".ckpt" + }, + { + "source_name": "manual_models", + "source": "/Volumes/2T/pymss/models/smoke/swin_upernet/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "dest": "/Volumes/2T/pymss/all_models/manual_models/smoke/swin_upernet/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "relpath": "manual_models/smoke/swin_upernet/model_swin_upernet_ep_56_sdr_10.6703.ckpt", + "size": 939903161, + "suffix": ".ckpt" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/1_HP-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/1_HP-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/1_HP-UVR.pth", + "size": 126792647, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/2_HP-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/2_HP-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/2_HP-UVR.pth", + "size": 126782699, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/3_HP-Vocal-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/3_HP-Vocal-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/3_HP-Vocal-UVR.pth", + "size": 126792647, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/4_HP-Vocal-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/4_HP-Vocal-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/4_HP-Vocal-UVR.pth", + "size": 126792647, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/5_HP-Karaoke-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/5_HP-Karaoke-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/5_HP-Karaoke-UVR.pth", + "size": 126782699, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/6_HP-Karaoke-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/6_HP-Karaoke-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/6_HP-Karaoke-UVR.pth", + "size": 126782699, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/7_HP2-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/7_HP2-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/7_HP2-UVR.pth", + "size": 550120123, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/8_HP2-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/8_HP2-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/8_HP2-UVR.pth", + "size": 550131079, + "suffix": ".pth" + }, + { + "source_name": "msst_webui_pretrain", + "source": "/Volumes/2T/pymss/MSST_WebUI_1.7.0_v2_cu128/pretrain/VR_Models/9_HP2-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/msst_webui_pretrain/VR_Models/9_HP2-UVR.pth", + "relpath": "msst_webui_pretrain/VR_Models/9_HP2-UVR.pth", + "size": 550131079, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/10_SP-UVR-2B-32000-1.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/10_SP-UVR-2B-32000-1.pth", + "relpath": "pymss_pretrain/VR_Models/10_SP-UVR-2B-32000-1.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/11_SP-UVR-2B-32000-2.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/11_SP-UVR-2B-32000-2.pth", + "relpath": "pymss_pretrain/VR_Models/11_SP-UVR-2B-32000-2.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/12_SP-UVR-3B-44100.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/12_SP-UVR-3B-44100.pth", + "relpath": "pymss_pretrain/VR_Models/12_SP-UVR-3B-44100.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/13_SP-UVR-4B-44100-1.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/13_SP-UVR-4B-44100-1.pth", + "relpath": "pymss_pretrain/VR_Models/13_SP-UVR-4B-44100-1.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/14_SP-UVR-4B-44100-2.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/14_SP-UVR-4B-44100-2.pth", + "relpath": "pymss_pretrain/VR_Models/14_SP-UVR-4B-44100-2.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/15_SP-UVR-MID-44100-1.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/15_SP-UVR-MID-44100-1.pth", + "relpath": "pymss_pretrain/VR_Models/15_SP-UVR-MID-44100-1.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/16_SP-UVR-MID-44100-2.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/16_SP-UVR-MID-44100-2.pth", + "relpath": "pymss_pretrain/VR_Models/16_SP-UVR-MID-44100-2.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/17_HP-Wind_Inst-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/17_HP-Wind_Inst-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/17_HP-Wind_Inst-UVR.pth", + "size": 223661285, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/1_HP-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/1_HP-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/1_HP-UVR.pth", + "size": 126792647, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/2_HP-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/2_HP-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/2_HP-UVR.pth", + "size": 126782699, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/3_HP-Vocal-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/3_HP-Vocal-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/3_HP-Vocal-UVR.pth", + "size": 126792647, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/4_HP-Vocal-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/4_HP-Vocal-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/4_HP-Vocal-UVR.pth", + "size": 126792647, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/5_HP-Karaoke-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/5_HP-Karaoke-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/5_HP-Karaoke-UVR.pth", + "size": 126782699, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/6_HP-Karaoke-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/6_HP-Karaoke-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/6_HP-Karaoke-UVR.pth", + "size": 126782699, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/7_HP2-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/7_HP2-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/7_HP2-UVR.pth", + "size": 550120123, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/8_HP2-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/8_HP2-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/8_HP2-UVR.pth", + "size": 550131079, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/9_HP2-UVR.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/9_HP2-UVR.pth", + "relpath": "pymss_pretrain/VR_Models/9_HP2-UVR.pth", + "size": 550131079, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/Harmonic_Noise_Separation_yxlllc.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/Harmonic_Noise_Separation_yxlllc.pth", + "relpath": "pymss_pretrain/VR_Models/Harmonic_Noise_Separation_yxlllc.pth", + "size": 59320122, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/MGM_HIGHEND_v4.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/MGM_HIGHEND_v4.pth", + "relpath": "pymss_pretrain/VR_Models/MGM_HIGHEND_v4.pth", + "size": 31861529, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/MGM_LOWEND_A_v4.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/MGM_LOWEND_A_v4.pth", + "relpath": "pymss_pretrain/VR_Models/MGM_LOWEND_A_v4.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/MGM_LOWEND_B_v4.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/MGM_LOWEND_B_v4.pth", + "relpath": "pymss_pretrain/VR_Models/MGM_LOWEND_B_v4.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/MGM_MAIN_v4.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/MGM_MAIN_v4.pth", + "relpath": "pymss_pretrain/VR_Models/MGM_MAIN_v4.pth", + "size": 31938951, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/UVR-BVE-4B_SN-44100-1.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/UVR-BVE-4B_SN-44100-1.pth", + "relpath": "pymss_pretrain/VR_Models/UVR-BVE-4B_SN-44100-1.pth", + "size": 223650277, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/UVR-De-Echo-Aggressive.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/UVR-De-Echo-Aggressive.pth", + "relpath": "pymss_pretrain/VR_Models/UVR-De-Echo-Aggressive.pth", + "size": 127139365, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/UVR-De-Echo-Normal.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/UVR-De-Echo-Normal.pth", + "relpath": "pymss_pretrain/VR_Models/UVR-De-Echo-Normal.pth", + "size": 127139365, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/UVR-DeEcho-DeReverb.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/UVR-DeEcho-DeReverb.pth", + "relpath": "pymss_pretrain/VR_Models/UVR-DeEcho-DeReverb.pth", + "size": 223650277, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/UVR-DeNoise-Lite.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/UVR-DeNoise-Lite.pth", + "relpath": "pymss_pretrain/VR_Models/UVR-DeNoise-Lite.pth", + "size": 17922277, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/UVR-DeNoise.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/UVR-DeNoise.pth", + "relpath": "pymss_pretrain/VR_Models/UVR-DeNoise.pth", + "size": 127139365, + "suffix": ".pth" + }, + { + "source_name": "pymss_pretrain", + "source": "/Volumes/2T/pymss/pretrain/VR_Models/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth", + "dest": "/Volumes/2T/pymss/all_models/pymss_pretrain/VR_Models/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth", + "relpath": "pymss_pretrain/VR_Models/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth", + "size": 58928133, + "suffix": ".pth" + } + ] +} diff --git a/README.md b/README.md new file mode 100644 index 0000000000000000000000000000000000000000..154df8298fab5ecf322016157858e08cd1bccbe1 --- /dev/null +++ b/README.md @@ -0,0 +1,3 @@ +--- +license: apache-2.0 +--- diff --git a/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.ckpt b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3ec76fce3a2089d4303238193e3a7f5343570e35 --- /dev/null +++ b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99d9af7f1ff20e63c393035513a655392818d66b4d7fc23d658175c1f15e8d76 +size 66541845 diff --git a/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.json b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.json new file mode 100644 index 0000000000000000000000000000000000000000..46ac04c72af03662c27935fc396265b131e7a48b --- /dev/null +++ b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.json @@ -0,0 +1,35 @@ +{ + "input": "models/smoke/apollo/Apollo_LQ_MP3_restoration.ckpt", + "output": "models/smoke/apollo/Apollo_LQ_MP3_restoration.pymss_state_dict.pt", + "top_level_type": "dict", + "top_level_keys": [ + "model_name", + "state_dict", + "model_args", + "infos" + ], + "state_dict_tensors": 654, + "state_dict_parameters": 16577708, + "first_keys": [ + "BN.0.0.weight", + "BN.0.1.weight", + "BN.0.1.bias", + "BN.1.0.weight", + "BN.1.1.weight", + "BN.1.1.bias", + "BN.2.0.weight", + "BN.2.1.weight", + "BN.2.1.bias", + "BN.3.0.weight", + "BN.3.1.weight", + "BN.3.1.bias", + "BN.4.0.weight", + "BN.4.1.weight", + "BN.4.1.bias", + "BN.5.0.weight", + "BN.5.1.weight", + "BN.5.1.bias", + "BN.6.0.weight", + "BN.6.1.weight" + ] +} \ No newline at end of file diff --git a/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.pt b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.pt new file mode 100644 index 0000000000000000000000000000000000000000..60295d77b35d1ddc6b9118b3db9080d67f83855c --- /dev/null +++ b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.pymss_state_dict.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6f3c0eaa983e319f9285668802115d67b7a1cbae6f8b522a82d322e006d537f +size 66528927 diff --git a/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.yaml b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.yaml new file mode 100644 index 0000000000000000000000000000000000000000..50d0bb0374325b1cf59384029cb0e687635de46a --- /dev/null +++ b/cleanup/audio_restoration/Apollo_LQ_MP3_restoration.yaml @@ -0,0 +1,31 @@ +audio: + chunk_size: 132300 + min_mean_abs: 0.0 + num_channels: 2 + sample_rate: 44100 +augmentations: + enable: false +inference: + batch_size: 1 + overlap_size: 99225 +model: + feature_dim: 256 + layer: 6 + sr: 44100 + win: 20 +training: + batch_size: 2 + coarse_loss_clip: true + grad_clip: 0 + instruments: + - restored + - addition + lr: 1.0 + num_epochs: 1000 + num_steps: 1000 + optimizer: prodigy + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: restored + use_amp: true diff --git a/cleanup/audio_restoration/apollo_model.ckpt b/cleanup/audio_restoration/apollo_model.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..00c435519ca6f1987e9bf1de0b70df2bfa0357c5 --- /dev/null +++ b/cleanup/audio_restoration/apollo_model.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61b93332da9cde13925b2726b0b31167b05bc20030271b19a583ccf76d69354b +size 66532080 diff --git a/cleanup/audio_restoration/apollo_model.yaml b/cleanup/audio_restoration/apollo_model.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8d58baa15b2f7da0c91317a04d548be67f462705 --- /dev/null +++ b/cleanup/audio_restoration/apollo_model.yaml @@ -0,0 +1,41 @@ +audio: + chunk_size: 132300 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 + +model: + sr: 44100 + win: 20 + feature_dim: 256 + layer: 6 + +training: + instruments: ['restored', 'addition'] + target_instrument: 'restored' + batch_size: 1 + gradient_accumulation_steps: 2 + num_steps: 1000 + num_epochs: 1000 + optimizer: 'prodigy' + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + coarse_loss_clip: true + grad_clip: 0 + q: 0.95 + use_amp: false + +optimizer: + weight_decay: 0.01 + betas: [0.5, 0.99] + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + +inference: + batch_size: 1 + overlap_size: 99225 diff --git a/cleanup/audio_restoration/apollo_model_v2.ckpt b/cleanup/audio_restoration/apollo_model_v2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..87e69f645ecf17bb470a62ec3c1cf55ba1630b1d --- /dev/null +++ b/cleanup/audio_restoration/apollo_model_v2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9255c2f43b467840260dc4c0f99f26bcdb0a7c3523d7b5165d13322ca955b41 +size 38159792 diff --git a/cleanup/audio_restoration/apollo_model_v2.yaml b/cleanup/audio_restoration/apollo_model_v2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8d58baa15b2f7da0c91317a04d548be67f462705 --- /dev/null +++ b/cleanup/audio_restoration/apollo_model_v2.yaml @@ -0,0 +1,41 @@ +audio: + chunk_size: 132300 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 + +model: + sr: 44100 + win: 20 + feature_dim: 256 + layer: 6 + +training: + instruments: ['restored', 'addition'] + target_instrument: 'restored' + batch_size: 1 + gradient_accumulation_steps: 2 + num_steps: 1000 + num_epochs: 1000 + optimizer: 'prodigy' + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + coarse_loss_clip: true + grad_clip: 0 + q: 0.95 + use_amp: false + +optimizer: + weight_decay: 0.01 + betas: [0.5, 0.99] + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + +inference: + batch_size: 1 + overlap_size: 99225 diff --git a/cleanup/audio_restoration/model_apollo_vocals_ep_54.ckpt b/cleanup/audio_restoration/model_apollo_vocals_ep_54.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..75fe20239b79f9113a68f8f580a7220c5041f140 --- /dev/null +++ b/cleanup/audio_restoration/model_apollo_vocals_ep_54.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59e1311f93e1f0fde6d5d11fa69d97e41cdee39be38f0cf4ccb80cfce34b2a2b +size 194278526 diff --git a/cleanup/audio_restoration/model_apollo_vocals_ep_54.yaml b/cleanup/audio_restoration/model_apollo_vocals_ep_54.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e3007186077dd689301d565caa159d94ffa3042b --- /dev/null +++ b/cleanup/audio_restoration/model_apollo_vocals_ep_54.yaml @@ -0,0 +1,31 @@ +audio: + chunk_size: 441000 + min_mean_abs: 0.0 + num_channels: 2 + sample_rate: 44100 +augmentations: + enable: false +inference: + batch_size: 1 + overlap_size: 330750 +model: + feature_dim: 384 + layer: 8 + sr: 44100 + win: 20 +training: + batch_size: 1 + coarse_loss_clip: true + grad_clip: 0 + instruments: + - restored + - addition + lr: 1.0 + num_epochs: 1000 + num_steps: 1000 + optimizer: prodigy + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: restored + use_amp: true diff --git a/cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt b/cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8d59dd9875ffadf10ca18f66aa3d20a0e289ce90 --- /dev/null +++ b/cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca8799531fe51c94172cc047226209ed48bf7d8c02e04671795a15d2a1c318af +size 913096801 diff --git a/cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.yaml b/cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6dedbda2e5595a7eb8f67f55815b1e7225bca423 --- /dev/null +++ b/cleanup/crowd_removal/mel_band_roformer_crowd_aufr33_viperx_sdr_8.7144.yaml @@ -0,0 +1,71 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - crowd + - other + lr: 1.0e-05 + patience: 8 + reduce_factor: 0.95 + target_instrument: crowd + num_epochs: 1000 + num_steps: 4032 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 17640 diff --git a/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3e232e1d5ebb6e2c74cf9803ca9fb4a3fb860d17 --- /dev/null +++ b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a25e3b233722cd81e2de7b8e798a3fef29d4b9799ccacda60b0dc958a1e2a5bb +size 913097300 diff --git a/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.yaml b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.yaml new file mode 100644 index 0000000000000000000000000000000000000000..418e6346640edc5393e9ea01716e365e4fbe2d58 --- /dev/null +++ b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.yaml @@ -0,0 +1,71 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - dry + - other + lr: 1.0e-05 + patience: 8 + reduce_factor: 0.95 + target_instrument: dry + num_epochs: 1000 + num_steps: 4032 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + +inference: + batch_size: 2 + dim_t: 256 + overlap_size: 17640 diff --git a/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8d1b634e263572a5a955577112e77036279c6287 --- /dev/null +++ b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c1c39191edc34e942ca7f2346ce6b6c0e1208a5f76349ffce6f696bd12910de +size 913097300 diff --git a/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.yaml b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.yaml new file mode 100644 index 0000000000000000000000000000000000000000..418e6346640edc5393e9ea01716e365e4fbe2d58 --- /dev/null +++ b/cleanup/noise_bleed_control/denoise_mel_band_roformer_aufr33_sdr_27.9959.yaml @@ -0,0 +1,71 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - dry + - other + lr: 1.0e-05 + patience: 8 + reduce_factor: 0.95 + target_instrument: dry + num_epochs: 1000 + num_steps: 4032 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + +inference: + batch_size: 2 + dim_t: 256 + overlap_size: 17640 diff --git a/cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.ckpt b/cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..eb05d0fc17641b53deabf68c1940dd237695026c --- /dev/null +++ b/cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c450bd66a98b49dd03231fc5ebb84121eef8418236b179423c2b171d62b04d9 +size 913101368 diff --git a/cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.yaml b/cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e02996a0b3711eb4e8174cfa9cad0d2faec6bae4 --- /dev/null +++ b/cleanup/noise_bleed_control/kimmel_unwa_ft2_bleedless.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.ckpt b/cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4bf56278572c098907e036b63e094bd7c63c1a20 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a9a9d10faa7f8997676a78e66d741d7acb9cc449334763f3c8f626d68ec6e575 +size 913102724 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.yaml b/cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4b2950dab657192d542a493702326e4f95d57c1c --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_bleed_suppressor_v1.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Bleed + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 24255 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.ckpt b/cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8b595aaf271c7d97cbca025436c370b5954e7786 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91aa7a546ed2e93482e4629c982d35b0d258bb3de6eeab497fd91658cc86c7fd +size 913026650 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.yaml b/cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_denoise_debleed_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..65745b5be537052b903ae5f13f4e26f51912f6db --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de972fb724601beef237abe94c8b934c73218e9baf3e344ab4c2333276e5cfe7 +size 913026650 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.yaml b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v1_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..881d22e5fc7fde06ca2c5de68277c4ab8bbceee7 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6109687febb8f18cd5a45207fee35f18ba8b9467b18a4b2e982a3b7dc04a9d72 +size 913026650 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.yaml b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v2_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..651d00bdc63039e222bd04f9e901d82044d7ff84 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5578545e094e584835b3184310ed1b12072f15d4b6ed8f4359ecc17358a66676 +size 913026650 diff --git a/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.yaml b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/cleanup/noise_bleed_control/mel_band_roformer_instrumental_bleedless_v3_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/BS-Roformer_LargeV1.ckpt b/experimental_general/BS-Roformer_LargeV1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4345482c2b8fc31b1e1c21fc46c5288833943e44 --- /dev/null +++ b/experimental_general/BS-Roformer_LargeV1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d1334565bb9e4c08f4b3891d5ac39e269b511baf66024a3fd8a10112d52b945 +size 740263433 diff --git a/experimental_general/BS-Roformer_LargeV1.yaml b/experimental_general/BS-Roformer_LargeV1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d4bdee9d5593db30d1af26e40bb7b93eff6a3741 --- /dev/null +++ b/experimental_general/BS-Roformer_LargeV1.yaml @@ -0,0 +1,169 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +augmentations: + all: + channel_shuffle: 0.5 + mp3_compression: 0.05 + mp3_compression_backend: lameenc + mp3_compression_max_bitrate: 320 + mp3_compression_min_bitrate: 32 + pedalboard_mp3_compressor: 0.005 + pedalboard_mp3_compressor_pedalboard_mp3_compressor_max: 9.999 + pedalboard_mp3_compressor_pedalboard_mp3_compressor_min: 0 + pedalboard_resample: 0.001 + pedalboard_resample_target_sample_rate_max: 44100 + pedalboard_resample_target_sample_rate_min: 4000 + random_inverse: 0.1 + random_polarity: 0.5 + vocals: + pedalboard_reverb: 0.01 + pedalboard_reverb_damping_max: 0.9 + pedalboard_reverb_damping_min: 0.1 + pedalboard_reverb_dry_level_max: 0.9 + pedalboard_reverb_dry_level_min: 0.1 + pedalboard_reverb_room_size_max: 0.9 + pedalboard_reverb_room_size_min: 0.1 + pedalboard_reverb_wet_level_max: 0.9 + pedalboard_reverb_wet_level_min: 0.1 + pedalboard_reverb_width_max: 1.0 + pedalboard_reverb_width_min: 0.9 + pitch_shift: 0.25 + pitch_shift_max_semitones: 7 + pitch_shift_min_semitones: -7 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_max_gain_db: 9 + seven_band_parametric_eq_min_gain_db: -9 + tanh_distortion: 0.01 + tanh_distortion_max: 0.7 + tanh_distortion_min: 0.1 + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 + mixup: false + mixup_loudness_max: 1.75 + mixup_loudness_min: 0.25 + mixup_probs: !!python/tuple + - 0.2 + - 0.02 +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 17640 +model: + attn_dropout: 0.1 + depth: 16 + dim: 512 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.1 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + batch_size: 1 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - vocals + - other + lr: 2.0e-07 + num_epochs: 1000 + num_steps: 5000 + optimizer: adamw8bit + other_fix: true + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: vocals + use_amp: true diff --git a/experimental_general/BS_Roformer_FNF_V1.ckpt b/experimental_general/BS_Roformer_FNF_V1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b9240f012a798e7ceb37b777e6b973f1c6a16e6e --- /dev/null +++ b/experimental_general/BS_Roformer_FNF_V1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10b6f24495595f976390083ba115371d29b0fd600400488342f168935da3e834 +size 613995746 diff --git a/experimental_general/BS_Roformer_FNF_V1.yaml b/experimental_general/BS_Roformer_FNF_V1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e11583e11b6275b285c8a1616310ca726743587a --- /dev/null +++ b/experimental_general/BS_Roformer_FNF_V1.yaml @@ -0,0 +1,140 @@ +audio: + chunk_size: 65536 + dim_f: 1024 + dim_t: 1536 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Voices + - Inst + patience: 10 + reduce_factor: 0.95 + target_instrument: Voices + num_epochs: 50 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1536 + overlap_size: 3277 + normalize: false \ No newline at end of file diff --git a/experimental_general/Fullness.ckpt b/experimental_general/Fullness.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8775859971808892b902b65d7d7976a8de3a0589 --- /dev/null +++ b/experimental_general/Fullness.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c67f47fa84146c8182f0061c194a51782439fe732e6ffecb810f0e4af7026f6d +size 913091027 diff --git a/experimental_general/Fullness.yaml b/experimental_general/Fullness.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f7faf965972bad76f477830309fce1af3a89ec65 --- /dev/null +++ b/experimental_general/Fullness.yaml @@ -0,0 +1,48 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: true +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/INSTV8.ckpt b/experimental_general/INSTV8.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..7a8649d903aaea0b08b72c476ea5bbc5f7ab55e1 --- /dev/null +++ b/experimental_general/INSTV8.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d51dae045064a5817cd7588c46bfe2f93542d4dbe66c1e1a6e75ca081299a53 +size 913026650 diff --git a/experimental_general/INSTV8.yaml b/experimental_general/INSTV8.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f7faf965972bad76f477830309fce1af3a89ec65 --- /dev/null +++ b/experimental_general/INSTV8.yaml @@ -0,0 +1,48 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: true +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/INSTV9.ckpt b/experimental_general/INSTV9.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..761017c5571aaa81bbb5a970df6097622ac2f9e9 --- /dev/null +++ b/experimental_general/INSTV9.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b689ffa942c48353bbad9e8df7b96d61a24b5c50256b0052579b327552565c8 +size 913026650 diff --git a/experimental_general/INSTV9.yaml b/experimental_general/INSTV9.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f7faf965972bad76f477830309fce1af3a89ec65 --- /dev/null +++ b/experimental_general/INSTV9.yaml @@ -0,0 +1,48 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: true +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/MelBandRoformerBigSYHFTV1.ckpt b/experimental_general/MelBandRoformerBigSYHFTV1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3b07b85890712d20ad115ead701bf7799c3b9928 --- /dev/null +++ b/experimental_general/MelBandRoformerBigSYHFTV1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2327e3e81f19e67c307f8c830c54267c09ecb0e9c6ad2b40a80c310899c955f +size 1479738496 diff --git a/experimental_general/MelBandRoformerBigSYHFTV1.yaml b/experimental_general/MelBandRoformerBigSYHFTV1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..13234b7ac647da60e5e73bc65e70a99383f30b46 --- /dev/null +++ b/experimental_general/MelBandRoformerBigSYHFTV1.yaml @@ -0,0 +1,48 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + instruments: + - vocals + - other + target_instrument: vocals + use_amp: true +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/MelBandRoformerSYHFT.ckpt b/experimental_general/MelBandRoformerSYHFT.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fc8e114879a3761ce9e6f901cab7b0adbc7035b5 --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFT.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f319dfcde4396ea3106658f457f5eb0bc577e113491f61ae8bab216fe84b0c0c +size 913096702 diff --git a/experimental_general/MelBandRoformerSYHFT.yaml b/experimental_general/MelBandRoformerSYHFT.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12963adb5177c57750fc603e8e4326d2bde04aa7 --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFT.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - vocals + - other + lr: 0.0001 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 100 + augmentation: true + augmentation_type: null + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: false + use_amp: true +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 17640 diff --git a/experimental_general/MelBandRoformerSYHFTB1.ckpt b/experimental_general/MelBandRoformerSYHFTB1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e318de8d91bc2d1e2338cf0aa621797c5a7449a6 --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTB1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:739df364e7e07d072e60850c91bc525b2f8a7de468b29dcc8a948ac67cd754f3 +size 913106158 diff --git a/experimental_general/MelBandRoformerSYHFTB1.yaml b/experimental_general/MelBandRoformerSYHFTB1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1715f6ab3a7992dedb7386a5da47a276603f76fe --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTB1.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false + augmentation_type: true + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0.35 + augmentation_loudness_max: 1 + q: 0.99 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + other_fix: true + use_amp: true +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/MelBandRoformerSYHFTV2.5.ckpt b/experimental_general/MelBandRoformerSYHFTV2.5.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..232404ba4a0d4e6d32b6f683711c58cd73b10c18 --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTV2.5.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:916e3a2c1e63b1457bcad823b98ca705e4933deffd2a5ab3a370e10f68bf47e2 +size 913090472 diff --git a/experimental_general/MelBandRoformerSYHFTV2.5.yaml b/experimental_general/MelBandRoformerSYHFTV2.5.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12963adb5177c57750fc603e8e4326d2bde04aa7 --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTV2.5.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - vocals + - other + lr: 0.0001 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 100 + augmentation: true + augmentation_type: null + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: false + use_amp: true +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 17640 diff --git a/experimental_general/MelBandRoformerSYHFTV2.ckpt b/experimental_general/MelBandRoformerSYHFTV2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6b34c66f158e43e0f2c11b6df91a040c5a11a23c --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTV2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e99f8efa5315300c197295592bd7e56c21c1d77e1884c904b5128c54a2a4632 +size 913095346 diff --git a/experimental_general/MelBandRoformerSYHFTV2.yaml b/experimental_general/MelBandRoformerSYHFTV2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12963adb5177c57750fc603e8e4326d2bde04aa7 --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTV2.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - vocals + - other + lr: 0.0001 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 100 + augmentation: true + augmentation_type: null + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: false + use_amp: true +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 17640 diff --git a/experimental_general/MelBandRoformerSYHFTV3Epsilon.ckpt b/experimental_general/MelBandRoformerSYHFTV3Epsilon.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..650b62bfd05945493a6529898ca0d0023ee7637d --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTV3Epsilon.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c886092e4aae13aa089263a0d54d483643f58c16ec221aed37268e2c1031397 +size 913090472 diff --git a/experimental_general/MelBandRoformerSYHFTV3Epsilon.yaml b/experimental_general/MelBandRoformerSYHFTV3Epsilon.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12963adb5177c57750fc603e8e4326d2bde04aa7 --- /dev/null +++ b/experimental_general/MelBandRoformerSYHFTV3Epsilon.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - vocals + - other + lr: 0.0001 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 100 + augmentation: true + augmentation_type: null + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: false + use_amp: true +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 17640 diff --git a/experimental_general/apollo_model_uni.ckpt b/experimental_general/apollo_model_uni.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b04eda5a4eb4bc3927c0b6e888b8c545fcbdbe19 --- /dev/null +++ b/experimental_general/apollo_model_uni.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11433256eca90143d32f1108d156ac756fd0201fbc69e5f736def4062fb52186 +size 146867952 diff --git a/experimental_general/apollo_model_uni.yaml b/experimental_general/apollo_model_uni.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ad81fc44a507fdf43d9994c0164a7e37f988095c --- /dev/null +++ b/experimental_general/apollo_model_uni.yaml @@ -0,0 +1,31 @@ +audio: + chunk_size: 882000 + min_mean_abs: 0.0 + num_channels: 2 + sample_rate: 44100 +augmentations: + enable: false +inference: + batch_size: 1 + overlap_size: 661500 +model: + feature_dim: 384 + layer: 6 + sr: 44100 + win: 20 +training: + batch_size: 2 + coarse_loss_clip: true + grad_clip: 0 + instruments: + - restored + - addition + lr: 1.0 + num_epochs: 1000 + num_steps: 1000 + optimizer: prodigy + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: restored + use_amp: true diff --git a/experimental_general/bs_roformer_anvuew_sdr_12.45.ckpt b/experimental_general/bs_roformer_anvuew_sdr_12.45.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..82739cd98737aa146407d6e2a410ba2104119a97 --- /dev/null +++ b/experimental_general/bs_roformer_anvuew_sdr_12.45.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d0f72ad0ac4154f5f4c3cdf230880bbb9a5bd01dc610241a2abfe787f0d7784 +size 204485563 diff --git a/experimental_general/bs_roformer_anvuew_sdr_12.45.yaml b/experimental_general/bs_roformer_anvuew_sdr_12.45.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cfc22a9d93e4dcb5e496f95bf8d7108aeeb53cb5 --- /dev/null +++ b/experimental_general/bs_roformer_anvuew_sdr_12.45.yaml @@ -0,0 +1,126 @@ +audio: + chunk_size: 960000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0001 +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: true + skip_connection: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - instrument + lr: 1.0e-05 + patience: 5 + reduce_factor: 0.9 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false + use_amp: true +inference: + batch_size: 2 + dim_t: 1876 + overlap_size: 48000 diff --git a/experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.ckpt b/experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a111a538bc67921ae8b51e937717923571e5f8cc --- /dev/null +++ b/experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60347271e8493fdff28ef558c3b2297afb869a1f4462594ed548038264bec395 +size 204493312 diff --git a/experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.yaml b/experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cfc22a9d93e4dcb5e496f95bf8d7108aeeb53cb5 --- /dev/null +++ b/experimental_general/bs_roformer_ft1_anvuew_sdr_12.55.yaml @@ -0,0 +1,126 @@ +audio: + chunk_size: 960000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0001 +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: true + skip_connection: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - instrument + lr: 1.0e-05 + patience: 5 + reduce_factor: 0.9 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false + use_amp: true +inference: + batch_size: 2 + dim_t: 1876 + overlap_size: 48000 diff --git a/experimental_general/bs_roformer_mag_anvuew.ckpt b/experimental_general/bs_roformer_mag_anvuew.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a14ab98c21fadfeb13f0d6f24c6299caf9cc6893 --- /dev/null +++ b/experimental_general/bs_roformer_mag_anvuew.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16792de7af1a5d5293debec24d4566b13b645e6707776234409cce7697ea933d +size 204502577 diff --git a/experimental_general/bs_roformer_mag_anvuew.yaml b/experimental_general/bs_roformer_mag_anvuew.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b8b6ef34e3ae2a263dcb66a75bac82d520f51f5b --- /dev/null +++ b/experimental_general/bs_roformer_mag_anvuew.yaml @@ -0,0 +1,126 @@ +audio: + chunk_size: 640000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0001 +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: true + skip_connection: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - instrument + lr: 1.0e-05 + patience: 1 + reduce_factor: 0.75 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false + use_amp: true +inference: + batch_size: 2 + dim_t: 1251 + overlap_size: 32000 diff --git a/experimental_general/bs_roformer_revive.ckpt b/experimental_general/bs_roformer_revive.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5df342a5d9eb8dcdc04f5d04fbedef415a7835d3 --- /dev/null +++ b/experimental_general/bs_roformer_revive.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f1d7e4bfdfef07c6b2bc1d65283a7d03c3c38f8c7dbc8d729b785f93c8b8699a +size 639326600 diff --git a/experimental_general/bs_roformer_revive.yaml b/experimental_general/bs_roformer_revive.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a75976332adf58655f3a47459f9bc6c0fb457cc1 --- /dev/null +++ b/experimental_general/bs_roformer_revive.yaml @@ -0,0 +1,131 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false + augmentation_type: null + use_mp3_compress: false + augmentation_mix: false + augmentation_loudness: false + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true + use_amp: true +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/bs_roformer_revive2.ckpt b/experimental_general/bs_roformer_revive2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6826b2658d30a6b6aa5f3896311f7cd4c3160bb9 --- /dev/null +++ b/experimental_general/bs_roformer_revive2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58098850c882a7472dad39f99fb8040ce6eaafe671cfe9881d89aea276bbb5f5 +size 639326600 diff --git a/experimental_general/bs_roformer_revive2.yaml b/experimental_general/bs_roformer_revive2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a75976332adf58655f3a47459f9bc6c0fb457cc1 --- /dev/null +++ b/experimental_general/bs_roformer_revive2.yaml @@ -0,0 +1,131 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false + augmentation_type: null + use_mp3_compress: false + augmentation_mix: false + augmentation_loudness: false + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true + use_amp: true +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/bs_roformer_revive3e.ckpt b/experimental_general/bs_roformer_revive3e.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..56be9dd0e47d16ef81bb1be38566cf0a35c8ad43 --- /dev/null +++ b/experimental_general/bs_roformer_revive3e.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b0751b9a15c591407c3b77f08eb4ad3005e42e96051f3f2b39760f1130c467b +size 639326600 diff --git a/experimental_general/bs_roformer_revive3e.yaml b/experimental_general/bs_roformer_revive3e.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a75976332adf58655f3a47459f9bc6c0fb457cc1 --- /dev/null +++ b/experimental_general/bs_roformer_revive3e.yaml @@ -0,0 +1,131 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false + augmentation_type: null + use_mp3_compress: false + augmentation_mix: false + augmentation_loudness: false + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true + use_amp: true +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/checkpoint-cmn_state_dict.ckpt b/experimental_general/checkpoint-cmn_state_dict.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..680bbe1cb989b2aa5ea9f9dec69509f7d0d33308 --- /dev/null +++ b/experimental_general/checkpoint-cmn_state_dict.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4099d1e87130ec930e99fd56089062a6192a4141a5a2305d2e92ededd0688c1 +size 149097604 diff --git a/experimental_general/checkpoint-cmn_state_dict.yaml b/experimental_general/checkpoint-cmn_state_dict.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6b5813f7d6db294f2912c3878f585764ee361d99 --- /dev/null +++ b/experimental_general/checkpoint-cmn_state_dict.yaml @@ -0,0 +1,77 @@ +cls: Bandit + +audio: + chunk_size: 384000 + num_channels: 2 + sample_rate: 48000 + min_mean_abs: 0.000 + +kwargs: + in_channels: 1 + stems: ['speech', 'music', 'sfx'] + band_type: musical + n_bands: 64 + normalize_channel_independently: false + treat_channel_as_feature: true + n_sqm_modules: 8 + emb_dim: 128 + rnn_dim: 256 + bidirectional: true + rnn_type: "GRU" + mlp_dim: 512 + hidden_activation: "Tanh" + hidden_activation_kwargs: null + complex_mask: true + use_freq_weights: true + n_fft: 2048 + win_length: 2048 + hop_length: 512 + window_fn: "hann_window" + wkwargs: null + power: null + center: true + normalized: true + pad_mode: "reflect" + onesided: true + +training: + batch_size: 4 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - speech + - music + - sfx + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 288000 diff --git a/experimental_general/checkpoint-deu_state_dict.ckpt b/experimental_general/checkpoint-deu_state_dict.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6107a9e4d85c7e7006dcfc9d9813a4c6284659b5 --- /dev/null +++ b/experimental_general/checkpoint-deu_state_dict.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67cb1e2c8d8b34c0072237ce5f326d5f64ca3eb8fb2d5f8d90680b39903d8b8d +size 149097604 diff --git a/experimental_general/checkpoint-deu_state_dict.yaml b/experimental_general/checkpoint-deu_state_dict.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6b5813f7d6db294f2912c3878f585764ee361d99 --- /dev/null +++ b/experimental_general/checkpoint-deu_state_dict.yaml @@ -0,0 +1,77 @@ +cls: Bandit + +audio: + chunk_size: 384000 + num_channels: 2 + sample_rate: 48000 + min_mean_abs: 0.000 + +kwargs: + in_channels: 1 + stems: ['speech', 'music', 'sfx'] + band_type: musical + n_bands: 64 + normalize_channel_independently: false + treat_channel_as_feature: true + n_sqm_modules: 8 + emb_dim: 128 + rnn_dim: 256 + bidirectional: true + rnn_type: "GRU" + mlp_dim: 512 + hidden_activation: "Tanh" + hidden_activation_kwargs: null + complex_mask: true + use_freq_weights: true + n_fft: 2048 + win_length: 2048 + hop_length: 512 + window_fn: "hann_window" + wkwargs: null + power: null + center: true + normalized: true + pad_mode: "reflect" + onesided: true + +training: + batch_size: 4 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - speech + - music + - sfx + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 288000 diff --git a/experimental_general/checkpoint-eng_state_dict.ckpt b/experimental_general/checkpoint-eng_state_dict.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c1431bb13a04e8448a51850b9cab0a3ae96ec46b --- /dev/null +++ b/experimental_general/checkpoint-eng_state_dict.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ff55ca2699f043bf2451288fe6d4738dde59767b67e28b4f23a1dd33bf4e99a +size 149097604 diff --git a/experimental_general/checkpoint-eng_state_dict.yaml b/experimental_general/checkpoint-eng_state_dict.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6b5813f7d6db294f2912c3878f585764ee361d99 --- /dev/null +++ b/experimental_general/checkpoint-eng_state_dict.yaml @@ -0,0 +1,77 @@ +cls: Bandit + +audio: + chunk_size: 384000 + num_channels: 2 + sample_rate: 48000 + min_mean_abs: 0.000 + +kwargs: + in_channels: 1 + stems: ['speech', 'music', 'sfx'] + band_type: musical + n_bands: 64 + normalize_channel_independently: false + treat_channel_as_feature: true + n_sqm_modules: 8 + emb_dim: 128 + rnn_dim: 256 + bidirectional: true + rnn_type: "GRU" + mlp_dim: 512 + hidden_activation: "Tanh" + hidden_activation_kwargs: null + complex_mask: true + use_freq_weights: true + n_fft: 2048 + win_length: 2048 + hop_length: 512 + window_fn: "hann_window" + wkwargs: null + power: null + center: true + normalized: true + pad_mode: "reflect" + onesided: true + +training: + batch_size: 4 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - speech + - music + - sfx + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 288000 diff --git a/experimental_general/checkpoint-fao_state_dict.ckpt b/experimental_general/checkpoint-fao_state_dict.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8f52539048b75799a5eb71e46562bab9f924097b --- /dev/null +++ b/experimental_general/checkpoint-fao_state_dict.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bae55e12a54dea95ce7827f86b0fc15f2ebd2df625272ed9b9bba2677c60c4a5 +size 149097604 diff --git a/experimental_general/checkpoint-fao_state_dict.yaml b/experimental_general/checkpoint-fao_state_dict.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6b5813f7d6db294f2912c3878f585764ee361d99 --- /dev/null +++ b/experimental_general/checkpoint-fao_state_dict.yaml @@ -0,0 +1,77 @@ +cls: Bandit + +audio: + chunk_size: 384000 + num_channels: 2 + sample_rate: 48000 + min_mean_abs: 0.000 + +kwargs: + in_channels: 1 + stems: ['speech', 'music', 'sfx'] + band_type: musical + n_bands: 64 + normalize_channel_independently: false + treat_channel_as_feature: true + n_sqm_modules: 8 + emb_dim: 128 + rnn_dim: 256 + bidirectional: true + rnn_type: "GRU" + mlp_dim: 512 + hidden_activation: "Tanh" + hidden_activation_kwargs: null + complex_mask: true + use_freq_weights: true + n_fft: 2048 + win_length: 2048 + hop_length: 512 + window_fn: "hann_window" + wkwargs: null + power: null + center: true + normalized: true + pad_mode: "reflect" + onesided: true + +training: + batch_size: 4 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - speech + - music + - sfx + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 288000 diff --git a/experimental_general/checkpoint-fra_state_dict.ckpt b/experimental_general/checkpoint-fra_state_dict.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..dc1a5d957a36adc61d623cc8e63144e8aac97680 --- /dev/null +++ b/experimental_general/checkpoint-fra_state_dict.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f27f67f45263d2110179c9e57270dc59319fe4382de051956e7efff937672df +size 149097604 diff --git a/experimental_general/checkpoint-fra_state_dict.yaml b/experimental_general/checkpoint-fra_state_dict.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6b5813f7d6db294f2912c3878f585764ee361d99 --- /dev/null +++ b/experimental_general/checkpoint-fra_state_dict.yaml @@ -0,0 +1,77 @@ +cls: Bandit + +audio: + chunk_size: 384000 + num_channels: 2 + sample_rate: 48000 + min_mean_abs: 0.000 + +kwargs: + in_channels: 1 + stems: ['speech', 'music', 'sfx'] + band_type: musical + n_bands: 64 + normalize_channel_independently: false + treat_channel_as_feature: true + n_sqm_modules: 8 + emb_dim: 128 + rnn_dim: 256 + bidirectional: true + rnn_type: "GRU" + mlp_dim: 512 + hidden_activation: "Tanh" + hidden_activation_kwargs: null + complex_mask: true + use_freq_weights: true + n_fft: 2048 + win_length: 2048 + hop_length: 512 + window_fn: "hann_window" + wkwargs: null + power: null + center: true + normalized: true + pad_mode: "reflect" + onesided: true + +training: + batch_size: 4 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - speech + - music + - sfx + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 288000 diff --git a/experimental_general/checkpoint-spa_state_dict.ckpt b/experimental_general/checkpoint-spa_state_dict.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..bbe4f60864136ac226422fad4a44a62c9254b335 --- /dev/null +++ b/experimental_general/checkpoint-spa_state_dict.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ba2813404727c0ae9e7895a6dc757ff0bcb0e21dc7e4b9159ba04c3e91a3d70 +size 149097604 diff --git a/experimental_general/checkpoint-spa_state_dict.yaml b/experimental_general/checkpoint-spa_state_dict.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6b5813f7d6db294f2912c3878f585764ee361d99 --- /dev/null +++ b/experimental_general/checkpoint-spa_state_dict.yaml @@ -0,0 +1,77 @@ +cls: Bandit + +audio: + chunk_size: 384000 + num_channels: 2 + sample_rate: 48000 + min_mean_abs: 0.000 + +kwargs: + in_channels: 1 + stems: ['speech', 'music', 'sfx'] + band_type: musical + n_bands: 64 + normalize_channel_independently: false + treat_channel_as_feature: true + n_sqm_modules: 8 + emb_dim: 128 + rnn_dim: 256 + bidirectional: true + rnn_type: "GRU" + mlp_dim: 512 + hidden_activation: "Tanh" + hidden_activation_kwargs: null + complex_mask: true + use_freq_weights: true + n_fft: 2048 + win_length: 2048 + hop_length: 512 + window_fn: "hann_window" + wkwargs: null + power: null + center: true + normalized: true + pad_mode: "reflect" + onesided: true + +training: + batch_size: 4 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - speech + - music + - sfx + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 288000 diff --git a/experimental_general/mel_band_roformer_small.ckpt b/experimental_general/mel_band_roformer_small.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5db369eedfade9297c978a4242c17b1e2769a572 --- /dev/null +++ b/experimental_general/mel_band_roformer_small.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:20cc592a9dff7dc34a3fb0bf399cd68c950b03ff5f334725e74feb3d0186272f +size 202573672 diff --git a/experimental_general/mel_band_roformer_small.yaml b/experimental_general/mel_band_roformer_small.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fd79b0be47e58b65ae2cf1bf95105e2c85081fdf --- /dev/null +++ b/experimental_general/mel_band_roformer_small.yaml @@ -0,0 +1,49 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 1 +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: true +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/melband_roformer_small_v1.ckpt b/experimental_general/melband_roformer_small_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..bdcf1b6bd953981b87ab14dedf35078e2a5ada46 --- /dev/null +++ b/experimental_general/melband_roformer_small_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a1b9f3c32946b66781b82d5de856703d0fdf3359f047830c1373c0310dcb522 +size 202583890 diff --git a/experimental_general/melband_roformer_small_v1.yaml b/experimental_general/melband_roformer_small_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..afe446ed2ea3c4078a4bd0d4528bec1490228bca --- /dev/null +++ b/experimental_general/melband_roformer_small_v1.yaml @@ -0,0 +1,49 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 1 +training: + instruments: + - vocals + - other + target_instrument: vocals + use_amp: true +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/merge_syhft.ckpt b/experimental_general/merge_syhft.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..0577bf7ae61e5225d8314023aa6db50e1277dab5 --- /dev/null +++ b/experimental_general/merge_syhft.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e00f50365e5e87990f0f2ce4f9ae68ffc5dd82d2b37718a430bb046706265e00 +size 913036186 diff --git a/experimental_general/merge_syhft.yaml b/experimental_general/merge_syhft.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12963adb5177c57750fc603e8e4326d2bde04aa7 --- /dev/null +++ b/experimental_general/merge_syhft.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - vocals + - other + lr: 0.0001 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 100 + augmentation: true + augmentation_type: null + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: false + use_amp: true +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 17640 diff --git a/experimental_general/model2.ckpt b/experimental_general/model2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..215d48b92144a49bd3b28c2ca203106ea9d240ac --- /dev/null +++ b/experimental_general/model2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5942ddaa713ee92f660098a5241572b6f05e46c4014f8836ac2130b35dc03516 +size 913106158 diff --git a/experimental_general/model2.yaml b/experimental_general/model2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1715f6ab3a7992dedb7386a5da47a276603f76fe --- /dev/null +++ b/experimental_general/model2.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false + augmentation_type: true + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0.35 + augmentation_loudness_max: 1 + q: 0.99 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + other_fix: true + use_amp: true +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/model3.ckpt b/experimental_general/model3.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a0534cf029fd9c53eb8c960bd998b948706f2c7e --- /dev/null +++ b/experimental_general/model3.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b889c8efdf19e3332e5f29834545aaa2c0287d3125c08dfa00273170702ad2cb +size 913106158 diff --git a/experimental_general/model3.yaml b/experimental_general/model3.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1715f6ab3a7992dedb7386a5da47a276603f76fe --- /dev/null +++ b/experimental_general/model3.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false + augmentation_type: true + use_mp3_compress: false + augmentation_mix: true + augmentation_loudness: true + augmentation_loudness_type: 1 + augmentation_loudness_min: 0.35 + augmentation_loudness_max: 1 + q: 0.99 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + other_fix: true + use_amp: true +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 24255 diff --git a/experimental_general/model_bs_roformer_ep_937_sdr_10.5309.ckpt b/experimental_general/model_bs_roformer_ep_937_sdr_10.5309.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b30e4893bf1e4b198ea8005346b01efaa135c8b7 --- /dev/null +++ b/experimental_general/model_bs_roformer_ep_937_sdr_10.5309.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2e825a03bc908cb04dbd88eddeefbf5147dd1cf1f95cebf453d9dbfabec494b +size 393068365 diff --git a/experimental_general/model_bs_roformer_ep_937_sdr_10.5309.yaml b/experimental_general/model_bs_roformer_ep_937_sdr_10.5309.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d785aff68df655da08015351c450b96f8c2ac9d4 --- /dev/null +++ b/experimental_general/model_bs_roformer_ep_937_sdr_10.5309.yaml @@ -0,0 +1,138 @@ +audio: + chunk_size: 131584 + dim_f: 1024 + dim_t: 256 + hop_length: 512 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 4 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - No Drum-Bass + - Drum-Bass + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: No Drum-Bass + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 512 + overlap_size: 6579 diff --git a/experimental_general/model_jazz_scnet.ckpt b/experimental_general/model_jazz_scnet.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..98a2e8277640262b5f0818344552a75a6004bb90 --- /dev/null +++ b/experimental_general/model_jazz_scnet.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0db306f83f69afd1ec4d391d6d452e3b1a49862c68d7a8b41097368965b487ff +size 42458509 diff --git a/experimental_general/model_jazz_scnet.yaml b/experimental_general/model_jazz_scnet.yaml new file mode 100644 index 0000000000000000000000000000000000000000..883ef6c1f2c029c13a54d8cc47b38b059295f8d3 --- /dev/null +++ b/experimental_general/model_jazz_scnet.yaml @@ -0,0 +1,83 @@ +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - drums + - bass + - piano + - other + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 10 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - piano + - other + lr: 5.0e-04 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 363825 + normalize: true diff --git a/experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt b/experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1378b6e8a932eccf94d3ec42a6b6ec81e2f98985 --- /dev/null +++ b/experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49248d3e106e95f351b4a51942b87c00b6c3d3f6c0da11f6cf4d86b2d8ba55f9 +size 2739398173 diff --git a/experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.yaml b/experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.yaml new file mode 100644 index 0000000000000000000000000000000000000000..253ea4825f82bc416b2a9b8eff4ab0259f62c8bc --- /dev/null +++ b/experimental_general/model_mel_band_roformer_ep_3_sdr_14.8359.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - dry + - other + lr: 1.0e-05 + patience: 8 + reduce_factor: 0.95 + target_instrument: dry + num_epochs: 1000 + num_steps: 4032 + augmentation: false + augmentation_type: null + use_mp3_compress: false + augmentation_mix: false + augmentation_loudness: false + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: false + use_amp: true + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 17640 diff --git a/instrumental/Inst_ExperimentalV1.ckpt b/instrumental/Inst_ExperimentalV1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..244c4bc7d40ec40b25dbd938d6a731f9c22ff5bb --- /dev/null +++ b/instrumental/Inst_ExperimentalV1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb19b968287d31b6d229816342aa7f08f7e4010b7c61be2427fbdb0c04a8a020 +size 913026650 diff --git a/instrumental/Inst_ExperimentalV1.yaml b/instrumental/Inst_ExperimentalV1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/Inst_ExperimentalV1.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/Inst_Fv8.ckpt b/instrumental/Inst_Fv8.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c8c97b99411a2442ce7d0af8a4d4b7375c4f7558 --- /dev/null +++ b/instrumental/Inst_Fv8.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:026caebb240226f64b1611e623735bb14c042f772464292aa1a26e896a759cc5 +size 913027055 diff --git a/instrumental/Inst_Fv8.yaml b/instrumental/Inst_Fv8.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/Inst_Fv8.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/Inst_Fv9.ckpt b/instrumental/Inst_Fv9.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c3011a5d755aa84e99f77d1f733942f117e98530 --- /dev/null +++ b/instrumental/Inst_Fv9.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6c02cd154bb49d6d08044a8dc935a94ec26d09f2a0e6a03cc80a65d05a98ff2 +size 913090472 diff --git a/instrumental/Inst_Fv9.yaml b/instrumental/Inst_Fv9.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/Inst_Fv9.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/Inst_GaboxFv9.ckpt b/instrumental/Inst_GaboxFv9.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fb205052a2975a269286f47b0e6718c69b193da6 --- /dev/null +++ b/instrumental/Inst_GaboxFv9.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:068d29a54e71f69ed871ca410e09b9877e153ad1439e825bd093573ea4de5762 +size 913091027 diff --git a/instrumental/Inst_GaboxFv9.yaml b/instrumental/Inst_GaboxFv9.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/Inst_GaboxFv9.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/Inst_GaboxV7.ckpt b/instrumental/Inst_GaboxV7.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5f5aa9428c851e83a36c0432a3baf218aa6ca261 --- /dev/null +++ b/instrumental/Inst_GaboxV7.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e725a860176acb475d983a1ddd9c1a99a619c69cc9ceda808dd294d10db746a5 +size 913026650 diff --git a/instrumental/Inst_GaboxV7.yaml b/instrumental/Inst_GaboxV7.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/Inst_GaboxV7.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/Neo_InstVFX.ckpt b/instrumental/Neo_InstVFX.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..369e00d6d51b9962e3e74005bd026eeb5ee26f59 --- /dev/null +++ b/instrumental/Neo_InstVFX.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10d95b882c81c2b462d54f35c4c5b81ac6eee04f6e3746af95a79066010b012a +size 913090472 diff --git a/instrumental/Neo_InstVFX.yaml b/instrumental/Neo_InstVFX.yaml new file mode 100644 index 0000000000000000000000000000000000000000..96c60a12d6987eab0cb2327c59a9e56bc10c3158 --- /dev/null +++ b/instrumental/Neo_InstVFX.yaml @@ -0,0 +1,66 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - Instrumental + - Vocals + lr: 1.0e-05 + patience: 1000 + reduce_factor: 0.95 + target_instrument: Instrumental + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + use_torch_checkpoint: true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 24255 diff --git a/instrumental/bs_roformer_inst_hyperacev2.ckpt b/instrumental/bs_roformer_inst_hyperacev2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d641521632f14636b34b5bdeff850a78b4ac74a0 --- /dev/null +++ b/instrumental/bs_roformer_inst_hyperacev2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d61178ef966d2b4e9ad456ffbbc6fd5b2828df07a7af32931142c1a5ff1fe6f +size 288725646 diff --git a/instrumental/bs_roformer_inst_hyperacev2.yaml b/instrumental/bs_roformer_inst_hyperacev2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bf93b11baa9a27bf2d4ebf23fd8ace2f5912a503 --- /dev/null +++ b/instrumental/bs_roformer_inst_hyperacev2.yaml @@ -0,0 +1,126 @@ +audio: + chunk_size: 960000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0001 +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: true + skip_connection: false +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - instrument + lr: 1.0e-05 + patience: 5 + reduce_factor: 0.9 + target_instrument: instrument + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false + use_amp: true +inference: + batch_size: 2 + dim_t: 1876 + overlap_size: 48000 diff --git a/instrumental/essid_mel_inst.ckpt b/instrumental/essid_mel_inst.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5d3b8ac6ac25c8e31fc2548509f802f5dd544f53 --- /dev/null +++ b/instrumental/essid_mel_inst.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ce4ee5c6d57122d69fd74705c9abe009be24e1d74711d0907fdfbe4eff1260c +size 913106900 diff --git a/instrumental/essid_mel_inst.yaml b/instrumental/essid_mel_inst.yaml new file mode 100644 index 0000000000000000000000000000000000000000..624c129eec480aa1011885efe80754e0304fb384 --- /dev/null +++ b/instrumental/essid_mel_inst.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Instrumental + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/instrumental/essid_mel_inst_old.ckpt b/instrumental/essid_mel_inst_old.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..34c320266f810b8b3b01bbf50c574bd748316f49 --- /dev/null +++ b/instrumental/essid_mel_inst_old.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a90e6c282f3b5fc146b08748865766fa2d5b9c404551191e5efe13f1bade145d +size 913107578 diff --git a/instrumental/essid_mel_inst_old.yaml b/instrumental/essid_mel_inst_old.yaml new file mode 100644 index 0000000000000000000000000000000000000000..624c129eec480aa1011885efe80754e0304fb384 --- /dev/null +++ b/instrumental/essid_mel_inst_old.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Instrumental + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/instrumental/inst_gaboxFlowersV10.ckpt b/instrumental/inst_gaboxFlowersV10.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9648ffa1ea11a9d4a5c3e7500a180152bc0d5825 --- /dev/null +++ b/instrumental/inst_gaboxFlowersV10.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe3aa51927beb60ff345b8b672c11760fb8bdaa6039d8c61631416a5b4f7a047 +size 489571079 diff --git a/instrumental/inst_gaboxFlowersV10.yaml b/instrumental/inst_gaboxFlowersV10.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9b2dd0d2a375e58a6f9bbb8d81e0c3020905217d --- /dev/null +++ b/instrumental/inst_gaboxFlowersV10.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/last_mel_band_roformer.ckpt b/instrumental/last_mel_band_roformer.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3f105eb6b429fc9986673f48acfdb0ea7dff7d1b --- /dev/null +++ b/instrumental/last_mel_band_roformer.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25a20a3660cc609d25e57880ee2a52d6933efd10636bf849386aeef3ea3b7ca3 +size 2739327117 diff --git a/instrumental/last_mel_band_roformer.yaml b/instrumental/last_mel_band_roformer.yaml new file mode 100644 index 0000000000000000000000000000000000000000..624c129eec480aa1011885efe80754e0304fb384 --- /dev/null +++ b/instrumental/last_mel_band_roformer.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Instrumental + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/instrumental/mel_band_roformer_instrumental_2_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_2_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..7758aa5f0d19e21b8cb82550a8eddb7a7ab10e23 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_2_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e03ca459c339f88b7521c367c897d0c3f5362b38a6cdb96cb28e625ca0f9931e +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_2_gabox.yaml b/instrumental/mel_band_roformer_instrumental_2_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_2_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_3_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_3_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1596ed4c464e4346d614bc48bb64a996975bc445 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_3_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9ec9f299cf617bf6afe1c382f4b0761cd9bee78323da94889951812328e10fb +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_3_gabox.yaml b/instrumental/mel_band_roformer_instrumental_3_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_3_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_becruily.ckpt b/instrumental/mel_band_roformer_instrumental_becruily.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9aa6c9ab355e9dfc090251f9703e299b9da7893d --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_becruily.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8da6632a1c25efb1c9be783ce9ea367d226d4b918cd6c3717c8b1d7a396041d +size 913106900 diff --git a/instrumental/mel_band_roformer_instrumental_becruily.yaml b/instrumental/mel_band_roformer_instrumental_becruily.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a860bc26ed4dbbf1ce0baacbc9dd9198e94b7cca --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_becruily.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Instrumental + - Vocals + lr: 0.0005 + patience: 2 + reduce_factor: 0.95 + target_instrument: Instrumental + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5a49522bec839a399c6545d2f812bae2ff5f32a1 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f50296e913b9af3b5b3b961e92877ef0d4a74f9a433e796e89960c4c2b1abe53 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_noise_v4_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e73ab6f21e270a489e7310d9fe81a150cdfceb13 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31140eccf271d2a9e8a538b092b1f70dfd6471aa5ad163b22bccc758b9f38b62 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v1_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6d849f3079743541477aeb8748cd2dac3f05e0d0 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c446c34551333dd3d45b8d0708658a10f28c5e289f8ec27b5f0e22803681bef3 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v2_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c90caf360ea211abf7467c96c8bde8bf656fb521 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbb229209a8942d34664e19d2f4862e357ea3108a4e8c04b69aa0aba523a4481 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v3_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9be13d367583e58a02232c90a8a0fbeec8880b81 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b03d9bf0fc643dabf006dee38fd798140b4884c1b65c573b4488c1a2a876ad84 +size 913091027 diff --git a/instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fullness_v4_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fv7z_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fv7z_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9cd07358dc11187373fb9052aaee3b7434ee4269 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fv7z_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef229f0dcd370c1767e4292981c59e5248814da45f32bfacebcc0f28adaa30a1 +size 913031195 diff --git a/instrumental/mel_band_roformer_instrumental_fv7z_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fv7z_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fv7z_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fv8_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fv8_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b017d3834de1d9d48cafa54dc7807d19c2ea0880 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fv8_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50bfa127d21f419e0da89730867d28c7ac4484c9473e6f313d036bc8b429df80 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_fv8_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fv8_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fv8_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fv8b_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fv8b_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c740232a2e6bcdac7f26f69e39fffccf6a2e5261 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fv8b_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a28f79ae38b92caaeb37c5ccdf3912316b135414353ec0d5553d507e3de8ef7 +size 913027055 diff --git a/instrumental/mel_band_roformer_instrumental_fv8b_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fv8b_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fv8b_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_fvx_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_fvx_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..93fa162e3077a085662452f147e5fbe0bacd108a --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fvx_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:545ef13b0cdbac505818a38db98e09c54e7c03ea17b4e0c895a531bfa352fa59 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_fvx_gabox.yaml b/instrumental/mel_band_roformer_instrumental_fvx_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_fvx_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6ecff8779cfc8fa3961db8589050b44a43131203 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b3ad6bd8bed3aaaa4d9320ea2ca910d140196a2302186db1754f3a8d8e16fb1 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_gabox.yaml b/instrumental/mel_band_roformer_instrumental_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv5_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv5_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1a192beafebebb2f4785e076cb13aba8a074b8ce --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv5_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38e935cf1e97afcc1de84d0bdb87dd8090bad530fa0df28e707d16448e1d38e2 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv5_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv5_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv5_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv5n_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv5n_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5cea44b8d02d4a8379a8fef7d0dc7e3a920b6294 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv5n_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:175203923fac3e52ae00e7e37d41e8a7fef5020b6ee4e4144f4786daabc54b34 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv5n_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv5n_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv5n_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv6_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv6_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9e4e1ed19b5579f8914a2229b05a859b8cb0bf34 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv6_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:677951b8556a27abe32e39705640638826e78101fa901a51ad73d20522be6d25 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv6_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv6_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv6_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv6n_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv6n_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..edf54dd70e2304cc057d3055a46c02fbb79ef937 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv6n_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:802f3e5d183d7c4b50dea147c320e61634f5be6ff55fa899fdebeaf0f3cf7f42 +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv6n_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv6n_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv6n_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv7_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv7_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..96b9a5ec1eb7dddb73acd155f3962118dc4980ee --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv7_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b4f12d81fd7fb02f38609216d59f0e42b3dca655fd90ca275ab5321b3e4d9aa +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv7_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv7_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv7_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv7n_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv7n_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..0bb8eb969ef04c187caa14bc813e8c690d396036 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv7n_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0ca36af5d1314be46b56c8a53b6be02f98511fa5d7e3e196fd895755e65be3c +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv7n_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv7n_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv7n_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv8_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv8_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9802ecab0b9fa4c33af5610bb61b822076d77dcd --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv8_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d7d97418012e1d241853260d82330869c3e945ffdef9d7841fd90f5b24f20ff +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv8_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv8_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv8_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/mel_band_roformer_instrumental_instv8n_gabox.ckpt b/instrumental/mel_band_roformer_instrumental_instv8n_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..29483cf0cabdabb930587ec5c0be8070d73cc969 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv8n_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25920c876e601d4ccbf1684b19ecdac4b9fcfcc7f48d2c095d81040ec3fecbea +size 913026650 diff --git a/instrumental/mel_band_roformer_instrumental_instv8n_gabox.yaml b/instrumental/mel_band_roformer_instrumental_instv8n_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..695ef9f9f11f3fb7ac79afcd8f6b1ba65b9dc5f9 --- /dev/null +++ b/instrumental/mel_band_roformer_instrumental_instv8n_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Instrumental + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt b/instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6b58dcb93c7b74ff878b794348357f77c622f7d9 --- /dev/null +++ b/instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:12219998d4fcf208636472d48efc4a05fa42d3e3e68c5f7abfcb9fd8dac9b9d1 +size 2739362141 diff --git a/instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.yaml b/instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.yaml new file mode 100644 index 0000000000000000000000000000000000000000..624c129eec480aa1011885efe80754e0304fb384 --- /dev/null +++ b/instrumental/model_mel_band_roformer_ep_0_sdr_11.4805.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Instrumental + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt b/instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f96572993ab7171fa41789208ccd6fb35e6f2322 --- /dev/null +++ b/instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da1da5e35fd83f5abd77a313b91a59bd3d23c7d60d39a0b4a54893c7f6807c03 +size 913097300 diff --git a/instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.yaml b/instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b7edddb800646f60d203848c91d2f81f337949df --- /dev/null +++ b/instrumental/model_mel_band_roformer_ep_0_sdr_16.8230.yaml @@ -0,0 +1,48 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + instruments: + - Vocals + - Instrumental + target_instrument: Instrumental +inference: + batch_size: 1 + dim_t: 1101 + chunk_size: 352800 + overlap_size: 17640 diff --git a/karaoke/bs_karaoke_gabox_IS.ckpt b/karaoke/bs_karaoke_gabox_IS.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b823f1392f73b999ae0387834ad0737dd3715297 --- /dev/null +++ b/karaoke/bs_karaoke_gabox_IS.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db8357825398d4231031ad1ab4aa12a94bcaad8d67e8ce5e4b3c5b48fdee1d4f +size 204483448 diff --git a/karaoke/bs_karaoke_gabox_IS.yaml b/karaoke/bs_karaoke_gabox_IS.yaml new file mode 100644 index 0000000000000000000000000000000000000000..737bea1e47d00c234de25abf1947f0bc30258791 --- /dev/null +++ b/karaoke/bs_karaoke_gabox_IS.yaml @@ -0,0 +1,129 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: True + skip_connection: False + + +training: + batch_size: 1 + gradient_accumulation_steps: 999 + grad_clip: 1 + instruments: ['vocals', 'other'] + lr: 1.0e-5 + patience: 1000000 + reduce_factor: 0.75 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: Fira + other_fix: True # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + use_torch_checkpoint: True + +inference: + batch_size: 6 + dim_t: 1251 + overlap_size: 17640 diff --git a/karaoke/bs_roformer_karaoke_anvuew.ckpt b/karaoke/bs_roformer_karaoke_anvuew.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..ca2738a799877108f8e59770f06826941331a5e7 --- /dev/null +++ b/karaoke/bs_roformer_karaoke_anvuew.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:206d04757cb5f75ca3b55f8a0a48f5c26aa2351d4ff3c7adbfc9affa30ea3ae4 +size 204486925 diff --git a/karaoke/bs_roformer_karaoke_anvuew.yaml b/karaoke/bs_roformer_karaoke_anvuew.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9509cabac09cf8b1175f22ae372f2c4233892800 --- /dev/null +++ b/karaoke/bs_roformer_karaoke_anvuew.yaml @@ -0,0 +1,131 @@ +audio: + chunk_size: 640000 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: True + skip_connection: False + + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['Vocals', 'Instrumental'] + lr: 5.0e-5 + patience: 7 + reduce_factor: 0.75 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: False # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + + + + +inference: + batch_size: 2 + dim_t: 1251 + overlap_size: 32000 diff --git a/karaoke/bs_roformer_karaoke_frazer_becruily.ckpt b/karaoke/bs_roformer_karaoke_frazer_becruily.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..89ef86629edab3d7580712f0feb5bba8175fd3de --- /dev/null +++ b/karaoke/bs_roformer_karaoke_frazer_becruily.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb90ee24c1154d83fbcfd27e96182f19e061557cc6e4746953125e08c29389f9 +size 204436907 diff --git a/karaoke/bs_roformer_karaoke_frazer_becruily.yaml b/karaoke/bs_roformer_karaoke_frazer_becruily.yaml new file mode 100644 index 0000000000000000000000000000000000000000..100c581079422aac86c029404845f180e2d61305 --- /dev/null +++ b/karaoke/bs_roformer_karaoke_frazer_becruily.yaml @@ -0,0 +1,129 @@ +audio: + chunk_size: 882000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 2001 + overlap_size: 44100 + normalize: false diff --git a/karaoke/mel_band_karaoke_fusion_aggressive.ckpt b/karaoke/mel_band_karaoke_fusion_aggressive.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..ae98012b2e732161a906481dea349c22b4e1cf2a --- /dev/null +++ b/karaoke/mel_band_karaoke_fusion_aggressive.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a769b82198727e243ed95cecbfb14bec6e3ddf16ba7b5dc9ab3bc130bb2e88b8 +size 913091027 diff --git a/karaoke/mel_band_karaoke_fusion_aggressive.yaml b/karaoke/mel_band_karaoke_fusion_aggressive.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e0e02ccc734415922c672f433ed6e33aaf095bc1 --- /dev/null +++ b/karaoke/mel_band_karaoke_fusion_aggressive.yaml @@ -0,0 +1,83 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 411 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: true + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 150 + num_steps: 100 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: false + +lora: + r: 4 + lora_alpha: 8 # alpha / rank > 1 + lora_dropout: 0.01 + merge_weights: True + fan_in_fan_out: False + enable_lora: [True] + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/karaoke/mel_band_karaoke_fusion_standard.ckpt b/karaoke/mel_band_karaoke_fusion_standard.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b14cfcc380d1e954bbf85a9130d6c591a5956764 --- /dev/null +++ b/karaoke/mel_band_karaoke_fusion_standard.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b22f8715c73e5de48dd57230b46b6c08ddf90e45e6bd086e793330e12f3f2d0 +size 913091027 diff --git a/karaoke/mel_band_karaoke_fusion_standard.yaml b/karaoke/mel_band_karaoke_fusion_standard.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e0e02ccc734415922c672f433ed6e33aaf095bc1 --- /dev/null +++ b/karaoke/mel_band_karaoke_fusion_standard.yaml @@ -0,0 +1,83 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 411 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: true + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 150 + num_steps: 100 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: false + +lora: + r: 4 + lora_alpha: 8 # alpha / rank > 1 + lora_dropout: 0.01 + merge_weights: True + fan_in_fan_out: False + enable_lora: [True] + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/karaoke/mel_band_roformer_bve_gonza.ckpt b/karaoke/mel_band_roformer_bve_gonza.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a2eca39d03b4260a9c50b809e987471e4c614b07 --- /dev/null +++ b/karaoke/mel_band_roformer_bve_gonza.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e003bbb97ebef78c59cd6b46c5fc3d0f2303cd49f6fb98eb8b8f0f8075899ae3 +size 913090472 diff --git a/karaoke/mel_band_roformer_bve_gonza.yaml b/karaoke/mel_band_roformer_bve_gonza.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c6a12d6fba0c5ff6f1f9562f42ee199a23ad9035 --- /dev/null +++ b/karaoke/mel_band_roformer_bve_gonza.yaml @@ -0,0 +1,75 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 411 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: true + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Lead + - Back + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: Lead + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: false + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/karaoke/mel_band_roformer_karaoke_becruily.ckpt b/karaoke/mel_band_roformer_karaoke_becruily.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..ceaad63144a44e85b64617f842686673070bf5dc --- /dev/null +++ b/karaoke/mel_band_roformer_karaoke_becruily.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3aa262ac01df870b9fc033e9c7b6cad33fe04fc9c148b6c40841326a515a0e0 +size 1719139254 diff --git a/karaoke/mel_band_roformer_karaoke_becruily.yaml b/karaoke/mel_band_roformer_karaoke_becruily.yaml new file mode 100644 index 0000000000000000000000000000000000000000..43251c836e6b8202e3b6ef60553b9437644de282 --- /dev/null +++ b/karaoke/mel_band_roformer_karaoke_becruily.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.0005 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/karaoke/mel_band_roformer_karaoke_gabox.ckpt b/karaoke/mel_band_roformer_karaoke_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..69688626b9c8b8655e6eb549e7909f64ae862819 --- /dev/null +++ b/karaoke/mel_band_roformer_karaoke_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:303fc631e7aa587e9dc1e6ac4bb3667c6ba53aacb6b6a90abcfcf57935b92bd8 +size 913026650 diff --git a/karaoke/mel_band_roformer_karaoke_gabox.yaml b/karaoke/mel_band_roformer_karaoke_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ccfd2aad09cd598e904e57a5384b554c1c0baf5c --- /dev/null +++ b/karaoke/mel_band_roformer_karaoke_gabox.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.0005 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/karaoke/mel_band_roformer_karaoke_gabox_v2.ckpt b/karaoke/mel_band_roformer_karaoke_gabox_v2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f1f8bd05fe7cd5bc814cbc757048f09e5df8f195 --- /dev/null +++ b/karaoke/mel_band_roformer_karaoke_gabox_v2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec34be50327aeaf1a996c27977f5c30d1ac80c0076d69683d3e5184c31ea29d3 +size 913090472 diff --git a/karaoke/mel_band_roformer_karaoke_gabox_v2.yaml b/karaoke/mel_band_roformer_karaoke_gabox_v2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ccfd2aad09cd598e904e57a5384b554c1c0baf5c --- /dev/null +++ b/karaoke/mel_band_roformer_karaoke_gabox_v2.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.0005 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt b/karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..977a5c60ab478ab0b78ac5bc7e5296f9185f5549 --- /dev/null +++ b/karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1de20d459332fe8869aeb01327a31df0032262706e1365114e852dc271779813 +size 913096801 diff --git a/karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.yaml b/karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7d37b5522d9b43f614db8fe37c4e987798e8b4a6 --- /dev/null +++ b/karaoke/model_mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.yaml @@ -0,0 +1,68 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + min_mean_abs: 0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 17640 +model: + attn_dropout: 0 + depth: 6 + dim: 384 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0 + flash_attn: true + freq_transformer_depth: 1 + heads: 8 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_bands: 60 + num_stems: 1 + sample_rate: 44100 + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + augmentation: false + augmentation_loudness: false + augmentation_loudness_max: 0 + augmentation_loudness_min: 0 + augmentation_loudness_type: 1 + augmentation_mix: false + augmentation_type: null + batch_size: 4 + coarse_loss_clip: false + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - karaoke + - other + lr: 1.0e-05 + num_epochs: 1000 + num_steps: 2000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: karaoke + use_mp3_compress: false diff --git a/karaoke/small_karaoke_gaboxaufr.ckpt b/karaoke/small_karaoke_gaboxaufr.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..429da8d0c30d1462e9499d11d413a5326257860e --- /dev/null +++ b/karaoke/small_karaoke_gaboxaufr.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6182fc508c0791d5a2ee4121b51f5ae9115a1bdbd209a8688b0620f4fd17fdca +size 202574227 diff --git a/karaoke/small_karaoke_gaboxaufr.yaml b/karaoke/small_karaoke_gaboxaufr.yaml new file mode 100644 index 0000000000000000000000000000000000000000..982a725dae195f1b931748cb2817dc8ade6ff271 --- /dev/null +++ b/karaoke/small_karaoke_gaboxaufr.yaml @@ -0,0 +1,56 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + use_torch_checkpoint: true + skip_connection: False + +training: + instruments: + - Instrumental + - Vocals + target_instrument: Vocals + use_amp: True + + inference: + batch_size: 6 + dim_t: 1101 + +inference: + overlap_size: 17640 diff --git a/legacy_models/0d19c1c6-0f06f20e.th b/legacy_models/0d19c1c6-0f06f20e.th new file mode 100644 index 0000000000000000000000000000000000000000..dadcebb0f5a2beb40367834688f751c4289634bd --- /dev/null +++ b/legacy_models/0d19c1c6-0f06f20e.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f06f20ed6ddc8058fa72ccc4845f3a88916eff7d007b623924193de217bbcf4 +size 178048329 diff --git a/legacy_models/1ef250f1-592467ce.th b/legacy_models/1ef250f1-592467ce.th new file mode 100644 index 0000000000000000000000000000000000000000..f913ca14132baa329bafe8e879d955556dbe43da --- /dev/null +++ b/legacy_models/1ef250f1-592467ce.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:592467cef0b0d878bc66fe38c8456353ff435334331baab08e1dc82beeefd596 +size 267602121 diff --git a/legacy_models/75fc33f5-1941ce65.th b/legacy_models/75fc33f5-1941ce65.th new file mode 100644 index 0000000000000000000000000000000000000000..7401e71bfbf0266d8004647514c8019101a4590b --- /dev/null +++ b/legacy_models/75fc33f5-1941ce65.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1941ce654b11df4132b9f4eae408556b4c83fad6fe26b4bc0dbcb36b975befb3 +size 167407275 diff --git a/legacy_models/7d865c68-3d5dd56b.th b/legacy_models/7d865c68-3d5dd56b.th new file mode 100644 index 0000000000000000000000000000000000000000..29a3bbc32d4aa5fe87611734dbb4e4e16386f9d2 --- /dev/null +++ b/legacy_models/7d865c68-3d5dd56b.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d5dd56b5bc986f136dff98655ded22b2b033f465ccec7a28640a6b15fd71ed6 +size 167918783 diff --git a/legacy_models/7ecf8ec1-70f50cc9.th b/legacy_models/7ecf8ec1-70f50cc9.th new file mode 100644 index 0000000000000000000000000000000000000000..e711c6c99ec4bcd406590accf2e3ed559861673d --- /dev/null +++ b/legacy_models/7ecf8ec1-70f50cc9.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70f50cc947d08f32e6dd8e2b687d398fa5ef9e51d1bd7600e32205d1f44be6b9 +size 178048329 diff --git a/legacy_models/902315c2-b39ce9c9.th b/legacy_models/902315c2-b39ce9c9.th new file mode 100644 index 0000000000000000000000000000000000000000..5193661d2f83a3856c08a696fb2db92c37bf5b83 --- /dev/null +++ b/legacy_models/902315c2-b39ce9c9.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b39ce9c97b5603fbaee99ec59fe1d2db570c77a37356c9c85542d881517c3302 +size 167405611 diff --git a/legacy_models/9a6b4851-03af0aa6.th b/legacy_models/9a6b4851-03af0aa6.th new file mode 100644 index 0000000000000000000000000000000000000000..3d23dd61aedad298a375c6fd44c5b4d7f81adc23 --- /dev/null +++ b/legacy_models/9a6b4851-03af0aa6.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03af0aa64af2c4f0795659c265fdead238fab31c064a8a4b5b051c1b22c8ad48 +size 267602121 diff --git a/legacy_models/HTDemucs4_FT_other.th b/legacy_models/HTDemucs4_FT_other.th new file mode 100644 index 0000000000000000000000000000000000000000..dc5aeb5ce23edf05faa98f84e7f8df6e3d121f1f --- /dev/null +++ b/legacy_models/HTDemucs4_FT_other.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef3bcb9c8b40d14ae5d51b6db2587339cc12c6b77c0be151ce6d69002e087bf2 +size 84141271 diff --git a/legacy_models/HTDemucs4_FT_other.yaml b/legacy_models/HTDemucs4_FT_other.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cac91b35c6f74f25e84de93ae0fce0f49376b2d2 --- /dev/null +++ b/legacy_models/HTDemucs4_FT_other.yaml @@ -0,0 +1,102 @@ +audio: + chunk_size: 485100 + hop_length: 1024 + min_mean_abs: 0.001 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 +htdemucs: + bottom_channels: 512 + cac: true + channels: 48 + channels_time: null + context: 1 + context_enc: 0 + dconv_comp: 8 + dconv_depth: 2 + dconv_init: 1e-3 + dconv_mode: 3 + depth: 4 + emb_scale: 10 + emb_smooth: true + end_iters: 0 + freq_emb: 0.2 + growth: 2 + kernel_size: 8 + multi_freqs: [] + multi_freqs_depth: 3 + nfft: 4096 + norm_groups: 4 + norm_starts: 4 + num_subbands: 1 + rescale: 0.1 + rewrite: true + stride: 4 + t_auto_sparsity: false + t_cape_augment: true + t_cape_glob_loc_scale: + - 5000.0 + - 1.0 + - 1.4 + t_cape_mean_normalize: true + t_cross_first: false + t_dropout: 0.0 + t_emb: sin + t_gelu: true + t_global_window: 100 + t_group_norm: false + t_heads: 8 + t_hidden_scale: 4.0 + t_layer_scale: true + t_layers: 5 + t_lr: null + t_mask_random_seed: 42 + t_mask_type: diag + t_max_period: 10000.0 + t_max_positions: 10000 + t_norm_first: true + t_norm_in: true + t_norm_in_group: false + t_norm_out: true + t_sin_random_shift: 0 + t_sparse_attn_window: 400 + t_sparse_cross_attn: false + t_sparse_self_attn: false + t_sparsity: 0.95 + t_weight_decay: 0.0 + t_weight_pos_embed: 1.0 + time_stride: 2 + wiener_iters: 0 + wiener_residual: false +inference: + batch_size: 1 + overlap_size: 363825 +model: htdemucs +training: + batch_size: 8 + channels: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - drums + - bass + - other + - vocals + lr: 9.0e-05 + normalize: true + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + samplerate: 44100 + segment: 11 + shift: 1 + target_instrument: null + use_amp: true diff --git a/legacy_models/cfa93e08-61801ae1.th b/legacy_models/cfa93e08-61801ae1.th new file mode 100644 index 0000000000000000000000000000000000000000..ae63ff8217e90e0549f57cfc43e388e6b27e645f --- /dev/null +++ b/legacy_models/cfa93e08-61801ae1.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61801ae1567d606c97a9c3469e943ae306d0a873eeb60d623ae7cfc7042b3f68 +size 167399275 diff --git a/legacy_models/demucs-e07c671f.th b/legacy_models/demucs-e07c671f.th new file mode 100644 index 0000000000000000000000000000000000000000..c815f961dfd091478311782ab0cda288256e4b95 --- /dev/null +++ b/legacy_models/demucs-e07c671f.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e07c671f8da9f16450730df0e4815e97c07555f3dc0a06d236201c268b372e33 +size 1062738817 diff --git a/legacy_models/demucs-e07c671f.yaml b/legacy_models/demucs-e07c671f.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0995b75dd7149595388255185ab68a1a81ea9477 --- /dev/null +++ b/legacy_models/demucs-e07c671f.yaml @@ -0,0 +1,2 @@ +models: ['ebf34a2db'] +segment: 44 \ No newline at end of file diff --git a/legacy_models/demucs.th b/legacy_models/demucs.th new file mode 100644 index 0000000000000000000000000000000000000000..faf85b1eca8b00dfb1c22b9addd3a1a24f706385 --- /dev/null +++ b/legacy_models/demucs.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6c4148ba0dc92242d82d7b3f2af55c77bd7cb4ff1a0a3028a523986f36a3cfd +size 2594202563 diff --git a/legacy_models/demucs.yaml b/legacy_models/demucs.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0995b75dd7149595388255185ab68a1a81ea9477 --- /dev/null +++ b/legacy_models/demucs.yaml @@ -0,0 +1,2 @@ +models: ['ebf34a2db'] +segment: 44 \ No newline at end of file diff --git a/legacy_models/demucs48_hq-28a1282c.th b/legacy_models/demucs48_hq-28a1282c.th new file mode 100644 index 0000000000000000000000000000000000000000..3932ef5510af903d40b5699be989567459e3b4d3 --- /dev/null +++ b/legacy_models/demucs48_hq-28a1282c.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28a1282c89ced9b343d284f61ed32cc3d03ce3e9d356076539924189d94c4eb5 +size 597870849 diff --git a/legacy_models/demucs_extra-3646af93.th b/legacy_models/demucs_extra-3646af93.th new file mode 100644 index 0000000000000000000000000000000000000000..1ef99fc7f1d42262feabdec9174461dd3a61af35 --- /dev/null +++ b/legacy_models/demucs_extra-3646af93.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3646af93cb495138aefd44801dc792cbfdd6e31ebced741c5707ce4461fbc4c6 +size 1062738817 diff --git a/legacy_models/demucs_extra-3646af93.yaml b/legacy_models/demucs_extra-3646af93.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0995b75dd7149595388255185ab68a1a81ea9477 --- /dev/null +++ b/legacy_models/demucs_extra-3646af93.yaml @@ -0,0 +1,2 @@ +models: ['ebf34a2db'] +segment: 44 \ No newline at end of file diff --git a/legacy_models/demucs_extra.th b/legacy_models/demucs_extra.th new file mode 100644 index 0000000000000000000000000000000000000000..81c8e5092e83af410b9933571c5c94dc512724f2 --- /dev/null +++ b/legacy_models/demucs_extra.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3331bcc5d09ba1d791c3cf851970242b0bb229ce81dbada557b6d39e8c6a6a87 +size 2594202563 diff --git a/legacy_models/demucs_extra.yaml b/legacy_models/demucs_extra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0995b75dd7149595388255185ab68a1a81ea9477 --- /dev/null +++ b/legacy_models/demucs_extra.yaml @@ -0,0 +1,2 @@ +models: ['ebf34a2db'] +segment: 44 \ No newline at end of file diff --git a/legacy_models/demucs_unittest-09ebc15f.th b/legacy_models/demucs_unittest-09ebc15f.th new file mode 100644 index 0000000000000000000000000000000000000000..60353ff6c788eb5fbfdb41d880569f27d31f03dc --- /dev/null +++ b/legacy_models/demucs_unittest-09ebc15f.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09ebc15ff947a7a8ed10ed6d8a1210f8c33189ea3bb3d0202310cb8d12968157 +size 4194049 diff --git a/legacy_models/demucs_unittest-09ebc15f.yaml b/legacy_models/demucs_unittest-09ebc15f.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0995b75dd7149595388255185ab68a1a81ea9477 --- /dev/null +++ b/legacy_models/demucs_unittest-09ebc15f.yaml @@ -0,0 +1,2 @@ +models: ['ebf34a2db'] +segment: 44 \ No newline at end of file diff --git a/legacy_models/e51eebcc-c1b80bdd.th b/legacy_models/e51eebcc-c1b80bdd.th new file mode 100644 index 0000000000000000000000000000000000000000..a786c99190033721bbdb524823ae601a768eb012 --- /dev/null +++ b/legacy_models/e51eebcc-c1b80bdd.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c1b80bdd6de58274abf359e66822a76f49ce2b9f086fc5dc917ac14598e6bebf +size 167399275 diff --git a/legacy_models/ebf34a2db.th b/legacy_models/ebf34a2db.th new file mode 100644 index 0000000000000000000000000000000000000000..1b95ef445d00f2b9f43d61727c76351d8bb2213f --- /dev/null +++ b/legacy_models/ebf34a2db.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dbdd521017d706829716055df8a86e7ee5d05a590508488c82ffb551c6edb918 +size 167390507 diff --git a/legacy_models/fa0cb7f9-100d8bf4.th b/legacy_models/fa0cb7f9-100d8bf4.th new file mode 100644 index 0000000000000000000000000000000000000000..0d3bea1cb75cda1c081f00e829a2733a4e509e66 --- /dev/null +++ b/legacy_models/fa0cb7f9-100d8bf4.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:100d8bf4dc5e228edfd9bb7a2ec44ad9f44d8f0a845964d38ee8c11ea8ca1399 +size 167405675 diff --git a/legacy_models/light.th b/legacy_models/light.th new file mode 100644 index 0000000000000000000000000000000000000000..f807ba7d0a19b21ce9ed1bedee8914c19c7d259f --- /dev/null +++ b/legacy_models/light.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:79d1ee3c1541c729c552327756954340a1a46a11ce0009dea77dc583e4b6269c +size 1062727712 diff --git a/legacy_models/tasnet-beb46fac.th b/legacy_models/tasnet-beb46fac.th new file mode 100644 index 0000000000000000000000000000000000000000..6acabb4e84cce888081f9401698426b829ad1021 --- /dev/null +++ b/legacy_models/tasnet-beb46fac.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:beb46fac044130652e64e0316447478762d76e03596cabf1cf483f4ac7cfbb0a +size 44035981 diff --git a/legacy_models/tasnet.th b/legacy_models/tasnet.th new file mode 100644 index 0000000000000000000000000000000000000000..41c4c00dc7910003584dd5aa921856899f5922e0 --- /dev/null +++ b/legacy_models/tasnet.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be56693f6a5c4854b124f95bb9dd043f3167614898493738ab52e25648bec8a2 +size 43973949 diff --git a/legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.pth b/legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.pth new file mode 100644 index 0000000000000000000000000000000000000000..e8235eeed743f442d3d8c167598448ef09cabf21 --- /dev/null +++ b/legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56165bdfa0dd5df7930ff76652200dc287e1bfef622bba5c812c736dc22067fe +size 223650277 diff --git a/legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.yaml b/legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1a7b9edb64e11d85b30776bbd0237695c363dd6c --- /dev/null +++ b/legacy_vr/vr_backing_vocal/UVR-BVE-4B_SN-44100-1.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'n_bins': 672, 'unstable_bins': 8, 'stable_bins': 530, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'convert_channels': 'stereo_n', 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False, 'bins': 672} + nout: 64 + nout_lstm: 128 + +training: + target_instrument: null + instruments: ['Vocals', 'Instrumental'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.pth b/legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.pth new file mode 100644 index 0000000000000000000000000000000000000000..28a23d764ee4460b4d3d1fefbde7e15c9deaae95 --- /dev/null +++ b/legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bd1d79d9c5d1b17d20f96f8a9f8aff1b55a83014f70712446bf420c0188e0a0 +size 127139365 diff --git a/legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.yaml b/legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.yaml new file mode 100644 index 0000000000000000000000000000000000000000..907b9ef7df2f87ebd70b683bb20440305e55f4c8 --- /dev/null +++ b/legacy_vr/vr_deecho/UVR-De-Echo-Aggressive.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 672, 'unstable_bins': 8, 'reduction_bins': 530, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: 48 + nout_lstm: 128 + +training: + target_instrument: null + instruments: ['No Echo', 'Echo'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_deecho/UVR-De-Echo-Normal.pth b/legacy_vr/vr_deecho/UVR-De-Echo-Normal.pth new file mode 100644 index 0000000000000000000000000000000000000000..0ff05edc8f7b50b8911a6f179fe5a512aa924fda --- /dev/null +++ b/legacy_vr/vr_deecho/UVR-De-Echo-Normal.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b849dd575643b075c257fb7a96c2ef5a79d7a5e7df74a2b319ad47118f1ee769 +size 127139365 diff --git a/legacy_vr/vr_deecho/UVR-De-Echo-Normal.yaml b/legacy_vr/vr_deecho/UVR-De-Echo-Normal.yaml new file mode 100644 index 0000000000000000000000000000000000000000..907b9ef7df2f87ebd70b683bb20440305e55f4c8 --- /dev/null +++ b/legacy_vr/vr_deecho/UVR-De-Echo-Normal.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 672, 'unstable_bins': 8, 'reduction_bins': 530, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: 48 + nout_lstm: 128 + +training: + target_instrument: null + instruments: ['No Echo', 'Echo'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.pth b/legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.pth new file mode 100644 index 0000000000000000000000000000000000000000..39fa772ed9097501fd2759fddef376e27d7c3ca4 --- /dev/null +++ b/legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e644028ec82865dc0fe082bc6fea85a43f7c71cfe375caee2da2d154aa661ee7 +size 223650277 diff --git a/legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.yaml b/legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.yaml new file mode 100644 index 0000000000000000000000000000000000000000..52aed492fceac7f513e88eb8e7e83decb67e7dd3 --- /dev/null +++ b/legacy_vr/vr_deecho_dereverb/UVR-DeEcho-DeReverb.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 672, 'unstable_bins': 8, 'reduction_bins': 530, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['No Reverb', 'Reverb'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_denoise/UVR-DeNoise-Lite.pth b/legacy_vr/vr_denoise/UVR-DeNoise-Lite.pth new file mode 100644 index 0000000000000000000000000000000000000000..e6c8b8a29f6c427aec2b1a55b235e59b2838113c --- /dev/null +++ b/legacy_vr/vr_denoise/UVR-DeNoise-Lite.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0023492fe98c406817b5253965de19ede65d1c147db015a3a428f07602e99571 +size 17922277 diff --git a/legacy_vr/vr_denoise/UVR-DeNoise-Lite.yaml b/legacy_vr/vr_denoise/UVR-DeNoise-Lite.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4d4f86f161a087360ec9e1740389cfa62d0fae87 --- /dev/null +++ b/legacy_vr/vr_denoise/UVR-DeNoise-Lite.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 1024, 'unstable_bins': 0, 'reduction_bins': 0, 'band': {1: {'sr': 44100, 'hl': 1024, 'n_fft': 2048, 'crop_start': 0, 'crop_stop': 1024, 'hpf_start': -1, 'res_type': 'sinc_best'}}, 'sr': 44100, 'pre_filter_start': 1023, 'pre_filter_stop': 1024, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: 16 + nout_lstm: 128 + +training: + target_instrument: null + instruments: ['Noise', 'No Noise'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_denoise/UVR-DeNoise.pth b/legacy_vr/vr_denoise/UVR-DeNoise.pth new file mode 100644 index 0000000000000000000000000000000000000000..fbb3b042175d49dfb778a8f4c99caa3c9edf2098 --- /dev/null +++ b/legacy_vr/vr_denoise/UVR-DeNoise.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5addf43ece5bddd18da9f575a02d7ffdb32342414e6ad7ac8d1dd7a04138a628 +size 127139365 diff --git a/legacy_vr/vr_denoise/UVR-DeNoise.yaml b/legacy_vr/vr_denoise/UVR-DeNoise.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f8ecde9bc6314ead68561ff5bae2f06e56f1c30 --- /dev/null +++ b/legacy_vr/vr_denoise/UVR-DeNoise.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 672, 'unstable_bins': 8, 'reduction_bins': 530, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: 48 + nout_lstm: 128 + +training: + target_instrument: null + instruments: ['Noise', 'No Noise'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth b/legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth new file mode 100644 index 0000000000000000000000000000000000000000..fd229237a6cb4e4c4af6961555d8d24d63a542d0 --- /dev/null +++ b/legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fff726b0adf09f2eadc2151645557db81e1a01fe15d91f9bbbfa9b50a007f1fd +size 58928133 diff --git a/legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.yaml b/legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f973ea7a33f4cd40bf7a6d44fe25e5474280a121 --- /dev/null +++ b/legacy_vr/vr_dereverb/UVR-DeReverb-aufr33-jarredou_4band_v4_ms_fullband.yaml @@ -0,0 +1,77 @@ +model: + model_params: + n_bins: 896 + unstable_bins: 9 + stable_bins: 530 + band: + 1: + sr: 7350 + hl: 96 + n_fft: 768 + crop_start: 0 + crop_stop: 102 + lpf_start: 30 + lpf_stop: 62 + res_type: polyphase + convert_channels: mid_side + 2: + sr: 7350 + hl: 96 + n_fft: 384 + crop_start: 5 + crop_stop: 104 + hpf_start: 30 + hpf_stop: 14 + lpf_start: 37 + lpf_stop: 73 + res_type: polyphase + convert_channels: mid_side + 3: + sr: 14700 + hl: 192 + n_fft: 640 + crop_start: 20 + crop_stop: 259 + hpf_start: 58 + hpf_stop: 29 + lpf_start: 191 + lpf_stop: 262 + res_type: polyphase + convert_channels: mid_side + 4: + sr: 44100 + hl: 576 + n_fft: 1152 + crop_start: 119 + crop_stop: 575 + hpf_start: 157 + hpf_stop: 110 + res_type: kaiser_fast + convert_channels: mid_side + sr: 44100 + pre_filter_start: -1 + pre_filter_stop: -1 + mid_side: false + mid_side_b: false + mid_side_b2: false + stereo_w: false + stereo_n: false + reverse: false + bins: 896 + vr_model_param: 4band_v4_ms_fullband + nout: 32 + nout_lstm: 128 +training: + target_instrument: null + instruments: + - Dry + - Reverb + use_amp: false +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 +audio: + sample_rate: 44100 diff --git a/legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.pth b/legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.pth new file mode 100644 index 0000000000000000000000000000000000000000..eab8b59808fe973a8154d7fa622f82c57c34db74 --- /dev/null +++ b/legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4dd9f8259692f9ceb5b05d86fd53553ef582a1c65184f4cfff2e00cdfe33dd2 +size 59320122 diff --git a/legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.yaml b/legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6f383712ac30e55fb944f23465fa34d63b346d02 --- /dev/null +++ b/legacy_vr/vr_harmonic_noise/Harmonic_Noise_Separation_yxlllc.yaml @@ -0,0 +1,40 @@ +model: + model_params: + bins: 1024 + unstable_bins: 0 + reduction_bins: 0 + band: + 1: + sr: 44100 + hl: 1024 + n_fft: 2048 + crop_start: 0 + crop_stop: 1024 + hpf_start: -1 + res_type: sinc_best + sr: 44100 + pre_filter_start: 1023 + pre_filter_stop: 1024 + mid_side: false + mid_side_b: false + mid_side_b2: false + stereo_w: false + stereo_n: false + reverse: false + vr_model_param: 1band_sr44100_hl1024 + nout: null + nout_lstm: null +training: + target_instrument: null + instruments: + - No Aspiration + - Aspiration + use_amp: false +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 +audio: + sample_rate: 44100 diff --git a/legacy_vr/vr_hp2/7_HP2-UVR.pth b/legacy_vr/vr_hp2/7_HP2-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..d91fc5482a76af6413dacc2a601df8b2510cd2c6 --- /dev/null +++ b/legacy_vr/vr_hp2/7_HP2-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:545ee733b704a263bf6f75328c5ed14f0fb439dcda7a6e8f53c8cecb92877afe +size 550120123 diff --git a/legacy_vr/vr_hp2/7_HP2-UVR.yaml b/legacy_vr/vr_hp2/7_HP2-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..38a905cf632db04e4926dda3d0e4058713062676 --- /dev/null +++ b/legacy_vr/vr_hp2/7_HP2-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'mid_side_b2': True, 'bins': 640, 'unstable_bins': 7, 'reduction_bins': 565, 'band': {1: {'sr': 11025, 'hl': 108, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 187, 'lpf_start': 92, 'lpf_stop': 186, 'res_type': 'polyphase'}, 2: {'sr': 22050, 'hl': 216, 'n_fft': 768, 'crop_start': 0, 'crop_stop': 212, 'hpf_start': 68, 'hpf_stop': 34, 'lpf_start': 174, 'lpf_stop': 209, 'res_type': 'polyphase'}, 3: {'sr': 44100, 'hl': 432, 'n_fft': 640, 'crop_start': 66, 'crop_stop': 307, 'hpf_start': 86, 'hpf_stop': 72, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 639, 'pre_filter_stop': 640, 'mid_side': False, 'mid_side_b': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_hp2/8_HP2-UVR.pth b/legacy_vr/vr_hp2/8_HP2-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..72633ec2c1c41105756bdc46c208910e7f9045ef --- /dev/null +++ b/legacy_vr/vr_hp2/8_HP2-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e072299a87cf3fcdd36da6adc6d2837bed0c07a208775a7678c3d5deeea58f79 +size 550131079 diff --git a/legacy_vr/vr_hp2/8_HP2-UVR.yaml b/legacy_vr/vr_hp2/8_HP2-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1b97f62c708e82fe92e7d8899619c4b3796e4706 --- /dev/null +++ b/legacy_vr/vr_hp2/8_HP2-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 668, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 186, 'lpf_start': 37, 'lpf_stop': 73, 'res_type': 'polyphase'}, 2: {'sr': 11025, 'hl': 128, 'n_fft': 512, 'crop_start': 4, 'crop_stop': 185, 'hpf_start': 36, 'hpf_stop': 18, 'lpf_start': 93, 'lpf_stop': 185, 'res_type': 'polyphase'}, 3: {'sr': 22050, 'hl': 256, 'n_fft': 512, 'crop_start': 46, 'crop_stop': 186, 'hpf_start': 93, 'hpf_stop': 46, 'lpf_start': 164, 'lpf_stop': 186, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 121, 'crop_stop': 382, 'hpf_start': 138, 'hpf_stop': 123, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 740, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_hp2/9_HP2-UVR.pth b/legacy_vr/vr_hp2/9_HP2-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..344c0743d56b86c4e1f1dad48bcfffc026dc3c6e --- /dev/null +++ b/legacy_vr/vr_hp2/9_HP2-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a99d032ed8c9d59077a8a41f0ceccb279d59ec10c87102167a283151d25ad9a2 +size 550131079 diff --git a/legacy_vr/vr_hp2/9_HP2-UVR.yaml b/legacy_vr/vr_hp2/9_HP2-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1b97f62c708e82fe92e7d8899619c4b3796e4706 --- /dev/null +++ b/legacy_vr/vr_hp2/9_HP2-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 668, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 186, 'lpf_start': 37, 'lpf_stop': 73, 'res_type': 'polyphase'}, 2: {'sr': 11025, 'hl': 128, 'n_fft': 512, 'crop_start': 4, 'crop_stop': 185, 'hpf_start': 36, 'hpf_stop': 18, 'lpf_start': 93, 'lpf_stop': 185, 'res_type': 'polyphase'}, 3: {'sr': 22050, 'hl': 256, 'n_fft': 512, 'crop_start': 46, 'crop_stop': 186, 'hpf_start': 93, 'hpf_stop': 46, 'lpf_start': 164, 'lpf_stop': 186, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 121, 'crop_stop': 382, 'hpf_start': 138, 'hpf_stop': 123, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 740, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_instrumental_extraction/1_HP-UVR.pth b/legacy_vr/vr_instrumental_extraction/1_HP-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..e82f0ea280de5f35a32bf213f45ad5b2ab4c11aa --- /dev/null +++ b/legacy_vr/vr_instrumental_extraction/1_HP-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9c654305888fe09668078ade76962b32d3782848f158192da7403a650e935a6 +size 126792647 diff --git a/legacy_vr/vr_instrumental_extraction/1_HP-UVR.yaml b/legacy_vr/vr_instrumental_extraction/1_HP-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1b97f62c708e82fe92e7d8899619c4b3796e4706 --- /dev/null +++ b/legacy_vr/vr_instrumental_extraction/1_HP-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 668, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 186, 'lpf_start': 37, 'lpf_stop': 73, 'res_type': 'polyphase'}, 2: {'sr': 11025, 'hl': 128, 'n_fft': 512, 'crop_start': 4, 'crop_stop': 185, 'hpf_start': 36, 'hpf_stop': 18, 'lpf_start': 93, 'lpf_stop': 185, 'res_type': 'polyphase'}, 3: {'sr': 22050, 'hl': 256, 'n_fft': 512, 'crop_start': 46, 'crop_stop': 186, 'hpf_start': 93, 'hpf_stop': 46, 'lpf_start': 164, 'lpf_stop': 186, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 121, 'crop_stop': 382, 'hpf_start': 138, 'hpf_stop': 123, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 740, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_instrumental_extraction/2_HP-UVR.pth b/legacy_vr/vr_instrumental_extraction/2_HP-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..75a8b276a9094cbf4e0180175b49f533b358461a --- /dev/null +++ b/legacy_vr/vr_instrumental_extraction/2_HP-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26d6400d2d2e90811ba2381828c8c3519196097375af524c3d0edd969acf5356 +size 126782699 diff --git a/legacy_vr/vr_instrumental_extraction/2_HP-UVR.yaml b/legacy_vr/vr_instrumental_extraction/2_HP-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fd79b4d13a9940b6d848efc3f0fb195adc643d4c --- /dev/null +++ b/legacy_vr/vr_instrumental_extraction/2_HP-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 672, 'unstable_bins': 8, 'reduction_bins': 637, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.pth b/legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..e1a6d0d913c7875ea67d2a892fcec4d44a14f426 --- /dev/null +++ b/legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe00891defbb61f4261500af22f7624f1a3df8dc75fa3998d1aece02e6be4537 +size 126782699 diff --git a/legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.yaml b/legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1f4e7038b81db2ec93d556d2df814ea22bd493f7 --- /dev/null +++ b/legacy_vr/vr_karaoke/5_HP-Karaoke-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 672, 'unstable_bins': 8, 'reduction_bins': 637, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'convert_channels': 'stereo_n', 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.pth b/legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..b0ab8fc7b1f85c340c608c4ffd43d964d1fb7637 --- /dev/null +++ b/legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ce7eaaa9e56f09366b788aebf6d3a72aec8145692c56f1e090e4e7e2d7ce65f +size 126782699 diff --git a/legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.yaml b/legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..38a905cf632db04e4926dda3d0e4058713062676 --- /dev/null +++ b/legacy_vr/vr_karaoke/6_HP-Karaoke-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'mid_side_b2': True, 'bins': 640, 'unstable_bins': 7, 'reduction_bins': 565, 'band': {1: {'sr': 11025, 'hl': 108, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 187, 'lpf_start': 92, 'lpf_stop': 186, 'res_type': 'polyphase'}, 2: {'sr': 22050, 'hl': 216, 'n_fft': 768, 'crop_start': 0, 'crop_stop': 212, 'hpf_start': 68, 'hpf_stop': 34, 'lpf_start': 174, 'lpf_stop': 209, 'res_type': 'polyphase'}, 3: {'sr': 44100, 'hl': 432, 'n_fft': 640, 'crop_start': 66, 'crop_stop': 307, 'hpf_start': 86, 'hpf_stop': 72, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 639, 'pre_filter_stop': 640, 'mid_side': False, 'mid_side_b': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_mgm/MGM_HIGHEND_v4.pth b/legacy_vr/vr_mgm/MGM_HIGHEND_v4.pth new file mode 100644 index 0000000000000000000000000000000000000000..eaac0cbcfb8eb02a574c6eea6936a45c88ad1163 --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_HIGHEND_v4.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67bf5241072ae938ae0e6bec44d2d8307d9db81a5299382a5cfe12188131f9ce +size 31861529 diff --git a/legacy_vr/vr_mgm/MGM_HIGHEND_v4.yaml b/legacy_vr/vr_mgm/MGM_HIGHEND_v4.yaml new file mode 100644 index 0000000000000000000000000000000000000000..583025cadb0abbaff8e6811be2e850f95da590f8 --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_HIGHEND_v4.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 1024, 'unstable_bins': 0, 'reduction_bins': 0, 'band': {1: {'sr': 44100, 'hl': 1024, 'n_fft': 2048, 'crop_start': 0, 'crop_stop': 1024, 'hpf_start': -1, 'res_type': 'sinc_best'}}, 'sr': 44100, 'pre_filter_start': 1023, 'pre_filter_stop': 1024, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_mgm/MGM_LOWEND_A_v4.pth b/legacy_vr/vr_mgm/MGM_LOWEND_A_v4.pth new file mode 100644 index 0000000000000000000000000000000000000000..60023486084acb358abf82b5a098ecd34cb03312 --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_LOWEND_A_v4.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5def7a35afecde33a77db214b3e9eeff0e47e8329ff5522d33332bc02b4b0e3 +size 31938951 diff --git a/legacy_vr/vr_mgm/MGM_LOWEND_A_v4.yaml b/legacy_vr/vr_mgm/MGM_LOWEND_A_v4.yaml new file mode 100644 index 0000000000000000000000000000000000000000..25de53bace387005c4872743455f8cd0837804fc --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_LOWEND_A_v4.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 1024, 'unstable_bins': 0, 'reduction_bins': 0, 'band': {1: {'sr': 32000, 'hl': 512, 'n_fft': 2048, 'crop_start': 0, 'crop_stop': 1024, 'hpf_start': -1, 'res_type': 'kaiser_fast'}}, 'sr': 32000, 'pre_filter_start': 1000, 'pre_filter_stop': 1021, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_mgm/MGM_LOWEND_B_v4.pth b/legacy_vr/vr_mgm/MGM_LOWEND_B_v4.pth new file mode 100644 index 0000000000000000000000000000000000000000..f1b963c0cb054a7135ce0cfdfff9e56354f44b71 --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_LOWEND_B_v4.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93d6a08abf6357e5f37d4eb6892267520f5bb9e90f45486b08a0857d5fcb2ec8 +size 31938951 diff --git a/legacy_vr/vr_mgm/MGM_LOWEND_B_v4.yaml b/legacy_vr/vr_mgm/MGM_LOWEND_B_v4.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9fdc0deffedc25692373beaccf9bdea53b93000b --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_LOWEND_B_v4.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 1024, 'unstable_bins': 0, 'reduction_bins': 0, 'band': {1: {'sr': 33075, 'hl': 384, 'n_fft': 2048, 'crop_start': 0, 'crop_stop': 1024, 'hpf_start': -1, 'res_type': 'sinc_best'}}, 'sr': 33075, 'pre_filter_start': 1000, 'pre_filter_stop': 1021, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_mgm/MGM_MAIN_v4.pth b/legacy_vr/vr_mgm/MGM_MAIN_v4.pth new file mode 100644 index 0000000000000000000000000000000000000000..b2e17497cf069f024054c58b3cc8fc9743593dd1 --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_MAIN_v4.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e6f0c0592333a3b215f61ac1e01f6c24c059f903f0789cf634e92daffae1dce +size 31938951 diff --git a/legacy_vr/vr_mgm/MGM_MAIN_v4.yaml b/legacy_vr/vr_mgm/MGM_MAIN_v4.yaml new file mode 100644 index 0000000000000000000000000000000000000000..41f0daaafe71c125130d1468ebd11024ffd79d24 --- /dev/null +++ b/legacy_vr/vr_mgm/MGM_MAIN_v4.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 1024, 'unstable_bins': 0, 'reduction_bins': 0, 'band': {1: {'sr': 44100, 'hl': 512, 'n_fft': 2048, 'crop_start': 0, 'crop_stop': 1024, 'hpf_start': -1, 'res_type': 'sinc_best'}}, 'sr': 44100, 'pre_filter_start': 1023, 'pre_filter_stop': 1024, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.pth b/legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.pth new file mode 100644 index 0000000000000000000000000000000000000000..f27302a3d3972a2fd74440c3c99ff5bba7cbd839 --- /dev/null +++ b/legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a178d8a8bc2c992bc5e5a045ef5ab8706d55c375d1bfe34700c1946d6c2d28b6 +size 31938951 diff --git a/legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.yaml b/legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..24a0d2460df6753380dcf6498de60ebfc331bf32 --- /dev/null +++ b/legacy_vr/vr_sp/10_SP-UVR-2B-32000-1.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 705, 'band': {1: {'sr': 6000, 'hl': 66, 'n_fft': 512, 'crop_start': 0, 'crop_stop': 240, 'lpf_start': 60, 'lpf_stop': 118, 'res_type': 'sinc_fastest'}, 2: {'sr': 32000, 'hl': 352, 'n_fft': 1024, 'crop_start': 22, 'crop_stop': 505, 'hpf_start': 44, 'hpf_stop': 23, 'res_type': 'sinc_medium'}}, 'sr': 32000, 'pre_filter_start': 710, 'pre_filter_stop': 731, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.pth b/legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.pth new file mode 100644 index 0000000000000000000000000000000000000000..27190898a57358cfba6fe0343e3432d6155204f2 --- /dev/null +++ b/legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e80e69ecd2244496710fa77e7d2c785bad93e2a87a7def65c38272cc9fd1613 +size 31938951 diff --git a/legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.yaml b/legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..24a0d2460df6753380dcf6498de60ebfc331bf32 --- /dev/null +++ b/legacy_vr/vr_sp/11_SP-UVR-2B-32000-2.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 705, 'band': {1: {'sr': 6000, 'hl': 66, 'n_fft': 512, 'crop_start': 0, 'crop_stop': 240, 'lpf_start': 60, 'lpf_stop': 118, 'res_type': 'sinc_fastest'}, 2: {'sr': 32000, 'hl': 352, 'n_fft': 1024, 'crop_start': 22, 'crop_stop': 505, 'hpf_start': 44, 'hpf_stop': 23, 'res_type': 'sinc_medium'}}, 'sr': 32000, 'pre_filter_start': 710, 'pre_filter_stop': 731, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_sp/12_SP-UVR-3B-44100.pth b/legacy_vr/vr_sp/12_SP-UVR-3B-44100.pth new file mode 100644 index 0000000000000000000000000000000000000000..fb5288aeaca78b613f29014c63e36f474d66c1ad --- /dev/null +++ b/legacy_vr/vr_sp/12_SP-UVR-3B-44100.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b826e6a44dbe007bde36c720bb2afeed008348cef1b61e35898fee5282d3918 +size 31938951 diff --git a/legacy_vr/vr_sp/12_SP-UVR-3B-44100.yaml b/legacy_vr/vr_sp/12_SP-UVR-3B-44100.yaml new file mode 100644 index 0000000000000000000000000000000000000000..91b013d369b83140d55751b94d576e10c5a4b2a1 --- /dev/null +++ b/legacy_vr/vr_sp/12_SP-UVR-3B-44100.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 5, 'reduction_bins': 733, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 768, 'crop_start': 0, 'crop_stop': 278, 'lpf_start': 28, 'lpf_stop': 140, 'res_type': 'polyphase'}, 2: {'sr': 22050, 'hl': 256, 'n_fft': 768, 'crop_start': 14, 'crop_stop': 322, 'hpf_start': 70, 'hpf_stop': 14, 'lpf_start': 283, 'lpf_stop': 314, 'res_type': 'polyphase'}, 3: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 131, 'crop_stop': 313, 'hpf_start': 154, 'hpf_stop': 141, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 757, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.pth b/legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.pth new file mode 100644 index 0000000000000000000000000000000000000000..8d0715db510d35c4964eab275b140ef761ff4f7e --- /dev/null +++ b/legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afb84b3d1da4a7e24033e5a0e85dbde2ac5f5e4b1e839788748735ac3bcc5046 +size 31938951 diff --git a/legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.yaml b/legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1b97f62c708e82fe92e7d8899619c4b3796e4706 --- /dev/null +++ b/legacy_vr/vr_sp/13_SP-UVR-4B-44100-1.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 668, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 186, 'lpf_start': 37, 'lpf_stop': 73, 'res_type': 'polyphase'}, 2: {'sr': 11025, 'hl': 128, 'n_fft': 512, 'crop_start': 4, 'crop_stop': 185, 'hpf_start': 36, 'hpf_stop': 18, 'lpf_start': 93, 'lpf_stop': 185, 'res_type': 'polyphase'}, 3: {'sr': 22050, 'hl': 256, 'n_fft': 512, 'crop_start': 46, 'crop_stop': 186, 'hpf_start': 93, 'hpf_stop': 46, 'lpf_start': 164, 'lpf_stop': 186, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 121, 'crop_stop': 382, 'hpf_start': 138, 'hpf_stop': 123, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 740, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.pth b/legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.pth new file mode 100644 index 0000000000000000000000000000000000000000..09ce4782c307d59cb4d29b9526dd39585485f0aa --- /dev/null +++ b/legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78404d85d6e4e11a3d2362f6577fc398f77406b7881fbac402e0c6191d6efca1 +size 31938951 diff --git a/legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.yaml b/legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1b97f62c708e82fe92e7d8899619c4b3796e4706 --- /dev/null +++ b/legacy_vr/vr_sp/14_SP-UVR-4B-44100-2.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 668, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 186, 'lpf_start': 37, 'lpf_stop': 73, 'res_type': 'polyphase'}, 2: {'sr': 11025, 'hl': 128, 'n_fft': 512, 'crop_start': 4, 'crop_stop': 185, 'hpf_start': 36, 'hpf_stop': 18, 'lpf_start': 93, 'lpf_stop': 185, 'res_type': 'polyphase'}, 3: {'sr': 22050, 'hl': 256, 'n_fft': 512, 'crop_start': 46, 'crop_stop': 186, 'hpf_start': 93, 'hpf_stop': 46, 'lpf_start': 164, 'lpf_stop': 186, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 121, 'crop_stop': 382, 'hpf_start': 138, 'hpf_stop': 123, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 740, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.pth b/legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.pth new file mode 100644 index 0000000000000000000000000000000000000000..77078902947cbfc26779d83ed895ac0c0dcbf57c --- /dev/null +++ b/legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd03d4061ce34569e8df9e0a13a235af7ff0d3c0c3def491db4439f9ff344dd0 +size 31938951 diff --git a/legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.yaml b/legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ff1b58cdc800193b799b7826262c77eed472790d --- /dev/null +++ b/legacy_vr/vr_sp/15_SP-UVR-MID-44100-1.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'mid_side': True, 'bins': 768, 'unstable_bins': 5, 'reduction_bins': 733, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 768, 'crop_start': 0, 'crop_stop': 278, 'lpf_start': 28, 'lpf_stop': 140, 'res_type': 'polyphase'}, 2: {'sr': 22050, 'hl': 256, 'n_fft': 768, 'crop_start': 14, 'crop_stop': 322, 'hpf_start': 70, 'hpf_stop': 14, 'lpf_start': 283, 'lpf_stop': 314, 'res_type': 'polyphase'}, 3: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 131, 'crop_stop': 313, 'hpf_start': 154, 'hpf_stop': 141, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 757, 'pre_filter_stop': 768, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.pth b/legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.pth new file mode 100644 index 0000000000000000000000000000000000000000..c2650722cdecf650037a9eb409fe7f0ea83c4026 --- /dev/null +++ b/legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0e5025987670f963a339db861b078d8526bf912f0ce2e00f732a7f0a3e846cc +size 31938951 diff --git a/legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.yaml b/legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ff1b58cdc800193b799b7826262c77eed472790d --- /dev/null +++ b/legacy_vr/vr_sp/16_SP-UVR-MID-44100-2.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'mid_side': True, 'bins': 768, 'unstable_bins': 5, 'reduction_bins': 733, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 768, 'crop_start': 0, 'crop_stop': 278, 'lpf_start': 28, 'lpf_stop': 140, 'res_type': 'polyphase'}, 2: {'sr': 22050, 'hl': 256, 'n_fft': 768, 'crop_start': 14, 'crop_stop': 322, 'hpf_start': 70, 'hpf_stop': 14, 'lpf_start': 283, 'lpf_stop': 314, 'res_type': 'polyphase'}, 3: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 131, 'crop_stop': 313, 'hpf_start': 154, 'hpf_stop': 141, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 757, 'pre_filter_stop': 768, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Instrumental', 'Vocals'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.pth b/legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..3a0e7a3081a92f32895d6478390e544488268f13 --- /dev/null +++ b/legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d3f9be265b2f2c3d684f23a9905674627c2552e0f6b5f36325cef8dc1f0ff6b +size 126792647 diff --git a/legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.yaml b/legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5660572cc107f4e67674ba781f12c3909cf26508 --- /dev/null +++ b/legacy_vr/vr_vocal_extraction/3_HP-Vocal-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 668, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 186, 'lpf_start': 37, 'lpf_stop': 73, 'res_type': 'polyphase'}, 2: {'sr': 11025, 'hl': 128, 'n_fft': 512, 'crop_start': 4, 'crop_stop': 185, 'hpf_start': 36, 'hpf_stop': 18, 'lpf_start': 93, 'lpf_stop': 185, 'res_type': 'polyphase'}, 3: {'sr': 22050, 'hl': 256, 'n_fft': 512, 'crop_start': 46, 'crop_stop': 186, 'hpf_start': 93, 'hpf_stop': 46, 'lpf_start': 164, 'lpf_stop': 186, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 121, 'crop_stop': 382, 'hpf_start': 138, 'hpf_stop': 123, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 740, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Vocals', 'Instrumental'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.pth b/legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..0a58d54b24fbabb2f8e3863ec7829b1fdaaab567 --- /dev/null +++ b/legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:295d1e1e3a52c17a040a4a6fe390214ad8ca0797f091b7c6d8db97d247a27271 +size 126792647 diff --git a/legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.yaml b/legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5660572cc107f4e67674ba781f12c3909cf26508 --- /dev/null +++ b/legacy_vr/vr_vocal_extraction/4_HP-Vocal-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 768, 'unstable_bins': 7, 'reduction_bins': 668, 'band': {1: {'sr': 11025, 'hl': 128, 'n_fft': 1024, 'crop_start': 0, 'crop_stop': 186, 'lpf_start': 37, 'lpf_stop': 73, 'res_type': 'polyphase'}, 2: {'sr': 11025, 'hl': 128, 'n_fft': 512, 'crop_start': 4, 'crop_stop': 185, 'hpf_start': 36, 'hpf_stop': 18, 'lpf_start': 93, 'lpf_stop': 185, 'res_type': 'polyphase'}, 3: {'sr': 22050, 'hl': 256, 'n_fft': 512, 'crop_start': 46, 'crop_stop': 186, 'hpf_start': 93, 'hpf_stop': 46, 'lpf_start': 164, 'lpf_stop': 186, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 512, 'n_fft': 768, 'crop_start': 121, 'crop_stop': 382, 'hpf_start': 138, 'hpf_stop': 123, 'res_type': 'sinc_medium'}}, 'sr': 44100, 'pre_filter_start': 740, 'pre_filter_stop': 768, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['Vocals', 'Instrumental'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.pth b/legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.pth new file mode 100644 index 0000000000000000000000000000000000000000..8833720c17aaf6c73b004e5bf5cb3be37bed2184 --- /dev/null +++ b/legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.pth @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:acc6d472b4b478da9c9ab5af45b167749e05a7f65b30c7d5988b3700a513aeee +size 223661285 diff --git a/legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.yaml b/legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a454680264e0bcfff05daaa5a3e944a8e6c7d4c4 --- /dev/null +++ b/legacy_vr/vr_wind_instrumental/17_HP-Wind_Inst-UVR.yaml @@ -0,0 +1,19 @@ +model: + model_params: {'bins': 672, 'unstable_bins': 8, 'reduction_bins': 530, 'band': {1: {'sr': 7350, 'hl': 80, 'n_fft': 640, 'crop_start': 0, 'crop_stop': 85, 'lpf_start': 25, 'lpf_stop': 53, 'res_type': 'polyphase'}, 2: {'sr': 7350, 'hl': 80, 'n_fft': 320, 'crop_start': 4, 'crop_stop': 87, 'hpf_start': 25, 'hpf_stop': 12, 'lpf_start': 31, 'lpf_stop': 62, 'res_type': 'polyphase'}, 3: {'sr': 14700, 'hl': 160, 'n_fft': 512, 'crop_start': 17, 'crop_stop': 216, 'hpf_start': 48, 'hpf_stop': 24, 'lpf_start': 139, 'lpf_stop': 210, 'res_type': 'polyphase'}, 4: {'sr': 44100, 'hl': 480, 'n_fft': 960, 'crop_start': 78, 'crop_stop': 383, 'hpf_start': 130, 'hpf_stop': 86, 'res_type': 'kaiser_fast'}}, 'sr': 44100, 'pre_filter_start': 668, 'pre_filter_stop': 672, 'mid_side': False, 'mid_side_b': False, 'mid_side_b2': False, 'stereo_w': False, 'stereo_n': False, 'reverse': False} + nout: null + nout_lstm: null + +training: + target_instrument: null + instruments: ['No Woodwinds', 'Woodwinds'] # [primary_stem, secondary_stem] + use_amp: false + +inference: + batch_size: 1 + aggression: 5 + high_end_process: false + post_process_threshold: 0.2 + window_size: 512 + +audio: + sample_rate: 44100 \ No newline at end of file diff --git a/music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.ckpt b/music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f0b50711c499ac1971506ec2254f172cd5f01ca1 --- /dev/null +++ b/music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c48284779f7d1258a6527d3aaa18a532d45c1f506e2dcc25d5ab179a8c5e2573 +size 148891175 diff --git a/music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.yaml b/music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a8873fe5e6611ed56e149885ded850da72c8e82b --- /dev/null +++ b/music_stems/dnr_stems/model_bandit_plus_dnr_sdr_11.47.yaml @@ -0,0 +1,78 @@ +name: "MultiMaskMultiSourceBandSplitRNN" +audio: + chunk_size: 264600 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + in_channel: 1 + stems: ['speech', 'music', 'effects'] + band_specs: "musical" + n_bands: 64 + fs: 44100 + require_no_overlap: false + require_no_gap: true + normalize_channel_independently: false + treat_channel_as_feature: true + n_sqm_modules: 8 + emb_dim: 128 + rnn_dim: 256 + bidirectional: true + rnn_type: "GRU" + mlp_dim: 512 + hidden_activation: "Tanh" + hidden_activation_kwargs: null + complex_mask: true + n_fft: 2048 + win_length: 2048 + hop_length: 512 + window_fn: "hann_window" + wkwargs: null + power: null + center: true + normalized: true + pad_mode: "constant" + onesided: true + +training: + batch_size: 4 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - Speech + - Music + - Effects + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 198450 diff --git a/music_stems/drum_stems/MDX23C-DrumSep-aufr33-jarredou.ckpt b/music_stems/drum_stems/MDX23C-DrumSep-aufr33-jarredou.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2f97198d19fcae11a05adbd6b2a746e1f466833f --- /dev/null +++ b/music_stems/drum_stems/MDX23C-DrumSep-aufr33-jarredou.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2a4aa53eb584d21eead358a4e66d1882ad182911be018f052b5da73be9096d0 +size 437652699 diff --git a/music_stems/drum_stems/MDX23C-DrumSep-aufr33-jarredou.yaml b/music_stems/drum_stems/MDX23C-DrumSep-aufr33-jarredou.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1730d7a76ab5eb8c4ad7f839dab1e7c5408279f4 --- /dev/null +++ b/music_stems/drum_stems/MDX23C-DrumSep-aufr33-jarredou.yaml @@ -0,0 +1,88 @@ +audio: + chunk_size: 130560 + dim_f: 1024 + dim_t: 256 + hop_length: 512 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + act: gelu + bottleneck_factor: 4 + growth: 128 + norm: InstanceNorm + num_blocks_per_scale: 2 + num_channels: 128 + num_scales: 5 + num_subbands: 4 + scale: + - 2 + - 2 + +training: + batch_size: 12 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - kick + - snare + - toms + - hh + - ride + - crash + lr: 9.0e-05 + patience: 30 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1268 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + # apply mp3 compression to mixture only (emulate downloading mp3 from internet) + mp3_compression_on_mixture: 0.0 + mp3_compression_on_mixture_bitrate_min: 32 + mp3_compression_on_mixture_bitrate_max: 320 + mp3_compression_on_mixture_backend: "lameenc" + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.01 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + mp3_compression: 0.0 + mp3_compression_min_bitrate: 32 + mp3_compression_max_bitrate: 320 + mp3_compression_backend: "lameenc" + pitch_shift: 0.1 + pitch_shift_min_semitones: -3 + pitch_shift_max_semitones: 3 + seven_band_parametric_eq: 0.5 + seven_band_parametric_eq_min_gain_db: -6 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.2 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + +inference: + batch_size: 1 + dim_t: 256 + num_overlap: 4 + overlap_size: 97920 \ No newline at end of file diff --git a/music_stems/drum_stems/gilliaan_drumsV1.ckpt b/music_stems/drum_stems/gilliaan_drumsV1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2120c887b7ffc45427d3ea5d46a20031ace35502 --- /dev/null +++ b/music_stems/drum_stems/gilliaan_drumsV1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f60c4370257257301922dec9c6528cfab1b82b02d772ae0a0090cb566d4c4ca7 +size 303490678 diff --git a/music_stems/drum_stems/gilliaan_drumsV1.yaml b/music_stems/drum_stems/gilliaan_drumsV1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b71cf82a4c7a580e2900cc579bd5004f9e120d34 --- /dev/null +++ b/music_stems/drum_stems/gilliaan_drumsV1.yaml @@ -0,0 +1,138 @@ +audio: + chunk_size: 588800 #882000 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'other'] + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 3000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 5.0e-6 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 29440 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/HTDemucs4.th b/music_stems/four_stem/HTDemucs4.th new file mode 100644 index 0000000000000000000000000000000000000000..94c5ce7eda92bb105307197a7be1d9635b417a1f --- /dev/null +++ b/music_stems/four_stem/HTDemucs4.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4 +size 84141911 diff --git a/music_stems/four_stem/HTDemucs4.yaml b/music_stems/four_stem/HTDemucs4.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cac91b35c6f74f25e84de93ae0fce0f49376b2d2 --- /dev/null +++ b/music_stems/four_stem/HTDemucs4.yaml @@ -0,0 +1,102 @@ +audio: + chunk_size: 485100 + hop_length: 1024 + min_mean_abs: 0.001 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 +htdemucs: + bottom_channels: 512 + cac: true + channels: 48 + channels_time: null + context: 1 + context_enc: 0 + dconv_comp: 8 + dconv_depth: 2 + dconv_init: 1e-3 + dconv_mode: 3 + depth: 4 + emb_scale: 10 + emb_smooth: true + end_iters: 0 + freq_emb: 0.2 + growth: 2 + kernel_size: 8 + multi_freqs: [] + multi_freqs_depth: 3 + nfft: 4096 + norm_groups: 4 + norm_starts: 4 + num_subbands: 1 + rescale: 0.1 + rewrite: true + stride: 4 + t_auto_sparsity: false + t_cape_augment: true + t_cape_glob_loc_scale: + - 5000.0 + - 1.0 + - 1.4 + t_cape_mean_normalize: true + t_cross_first: false + t_dropout: 0.0 + t_emb: sin + t_gelu: true + t_global_window: 100 + t_group_norm: false + t_heads: 8 + t_hidden_scale: 4.0 + t_layer_scale: true + t_layers: 5 + t_lr: null + t_mask_random_seed: 42 + t_mask_type: diag + t_max_period: 10000.0 + t_max_positions: 10000 + t_norm_first: true + t_norm_in: true + t_norm_in_group: false + t_norm_out: true + t_sin_random_shift: 0 + t_sparse_attn_window: 400 + t_sparse_cross_attn: false + t_sparse_self_attn: false + t_sparsity: 0.95 + t_weight_decay: 0.0 + t_weight_pos_embed: 1.0 + time_stride: 2 + wiener_iters: 0 + wiener_residual: false +inference: + batch_size: 1 + overlap_size: 363825 +model: htdemucs +training: + batch_size: 8 + channels: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - drums + - bass + - other + - vocals + lr: 9.0e-05 + normalize: true + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + samplerate: 44100 + segment: 11 + shift: 1 + target_instrument: null + use_amp: true diff --git a/music_stems/four_stem/MelBandRoformer4StemFTLarge.ckpt b/music_stems/four_stem/MelBandRoformer4StemFTLarge.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..79988facfb10e352f59e8983485a6bdfd5f636de --- /dev/null +++ b/music_stems/four_stem/MelBandRoformer4StemFTLarge.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:590358e6abe3255d87d7173ca2e4005de082f14cbb466ea4c5ffedbcc3964506 +size 3764030874 diff --git a/music_stems/four_stem/MelBandRoformer4StemFTLarge.yaml b/music_stems/four_stem/MelBandRoformer4StemFTLarge.yaml new file mode 100644 index 0000000000000000000000000000000000000000..650c002a96602d5df7dbcbe9c0d4710b23985fa2 --- /dev/null +++ b/music_stems/four_stem/MelBandRoformer4StemFTLarge.yaml @@ -0,0 +1,70 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 882 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 2049 + sample_rate: 44100 + stft_n_fft: 4096 + stft_hop_length: 882 + stft_win_length: 4096 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: false + skip_connection: true +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 + normalize: false + other_fix: false + use_amp: true +inference: + batch_size: 4 + dim_t: 1101 + normalize: false + overlap_size: 24255 diff --git a/music_stems/four_stem/SCNet-large_starrytong_fixed.ckpt b/music_stems/four_stem/SCNet-large_starrytong_fixed.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..ddaa50ee05e3ca837df5ec3d540f34443aa66af7 --- /dev/null +++ b/music_stems/four_stem/SCNet-large_starrytong_fixed.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65900dfa07d6b6e5d784c0f143920200a4bd281d6e78a806c549d0b912d5885e +size 168852258 diff --git a/music_stems/four_stem/SCNet-large_starrytong_fixed.yaml b/music_stems/four_stem/SCNet-large_starrytong_fixed.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b94b0f4db9f711e5258440a9996cb2bc501dfb72 --- /dev/null +++ b/music_stems/four_stem/SCNet-large_starrytong_fixed.yaml @@ -0,0 +1,88 @@ +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - drums + - bass + - other + - vocals + audio_channels: 2 + dims: + - 4 + - 64 + - 128 + - 256 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.225 + - 0.372 + - 0.403 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 6 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + # lr: 1.0e-04 + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 363825 + normalize: true diff --git a/music_stems/four_stem/bs_roformer_4stems_ft.ckpt b/music_stems/four_stem/bs_roformer_4stems_ft.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..776c9a51208964dd72564cd742d3416a60d02eac --- /dev/null +++ b/music_stems/four_stem/bs_roformer_4stems_ft.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a64c61c5972fe3fdf331d98f06bb4d6e237147c83a4bc71fd4e00e254c9b7652 +size 527245586 diff --git a/music_stems/four_stem/bs_roformer_4stems_ft.yaml b/music_stems/four_stem/bs_roformer_4stems_ft.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e1053a253308b0327fb42dd90fcbd2d07c6cfc15 --- /dev/null +++ b/music_stems/four_stem/bs_roformer_4stems_ft.yaml @@ -0,0 +1,191 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +augmentations: + all: + channel_shuffle: 0.5 + random_inverse: 0.1 + random_polarity: 0.5 + bass: + pitch_shift: 0.1 + pitch_shift_max_semitones: 2 + pitch_shift_min_semitones: -2 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_max_gain_db: 6 + seven_band_parametric_eq_min_gain_db: -3 + tanh_distortion: 0.1 + tanh_distortion_max: 0.5 + tanh_distortion_min: 0.1 + drums: + pitch_shift: 0.1 + pitch_shift_max_semitones: 5 + pitch_shift_min_semitones: -5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_max_gain_db: 9 + seven_band_parametric_eq_min_gain_db: -9 + tanh_distortion: 0.1 + tanh_distortion_max: 0.6 + tanh_distortion_min: 0.1 + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 + mixup: true + mixup_loudness_max: 1.5 + mixup_loudness_min: 0.5 + mixup_probs: !!python/tuple + - 0.2 + - 0.02 + other: + gaussian_noise: 0.1 + gaussian_noise_max_amplitude: 0.015 + gaussian_noise_min_amplitude: 0.001 + pitch_shift: 0.1 + pitch_shift_max_semitones: 4 + pitch_shift_min_semitones: -4 + time_stretch: 0.1 + time_stretch_max_rate: 1.25 + time_stretch_min_rate: 0.8 + vocals: + pitch_shift: 0.1 + pitch_shift_max_semitones: 5 + pitch_shift_min_semitones: -5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_max_gain_db: 9 + seven_band_parametric_eq_min_gain_db: -9 + tanh_distortion: 0.1 + tanh_distortion_max: 0.7 + tanh_distortion_min: 0.1 +inference: + batch_size: 2 + dim_t: 2048 + overlap_size: 24255 +model: + attn_dropout: 0.1 + depth: 8 + dim: 384 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.1 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 4 + skip_connection: false + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + augmentation: false + augmentation_loudness: true + augmentation_loudness_max: 1.5 + augmentation_loudness_min: 0.5 + augmentation_loudness_type: 1 + augmentation_mix: true + augmentation_type: simple1 + batch_size: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - drums + - bass + - other + - vocals + lr: 1.0e-05 + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 3 + q: 0.95 + reduce_factor: 0.95 + target_instrument: null + use_amp: true + use_mp3_compress: false diff --git a/music_stems/four_stem/mel_band_roformer_4stems_large_ver1.ckpt b/music_stems/four_stem/mel_band_roformer_4stems_large_ver1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2472823fc801f44ef77d64b29279dfb7389228fd --- /dev/null +++ b/music_stems/four_stem/mel_band_roformer_4stems_large_ver1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2ca1ac95e203d2630994c3d4a0cebabd7160e70bb010adab794843340fb704f +size 3764030874 diff --git a/music_stems/four_stem/mel_band_roformer_4stems_large_ver1.yaml b/music_stems/four_stem/mel_band_roformer_4stems_large_ver1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e75b942cb4edb03dfc0e019b96d28f4320df7f18 --- /dev/null +++ b/music_stems/four_stem/mel_band_roformer_4stems_large_ver1.yaml @@ -0,0 +1,162 @@ +audio: + chunk_size: 661500 + dim_f: 1024 + dim_t: 1101 + hop_length: 882 + n_fft: 4096 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0001 +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 2049 + sample_rate: 44100 + stft_n_fft: 4096 + stft_hop_length: 882 + stft_win_length: 4096 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: false + skip_connection: true +training: + batch_size: 1 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + lr: 2.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 300 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + read_metadata_procs: 8 + normalize: false + other_fix: false + use_amp: true +augmentations: + enable: false + loudness: false + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true + mixup_probs: !!python/tuple + - 0.2 + - 0.02 + - 0.002 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + mp3_compression_on_mixture: 0.01 + mp3_compression_on_mixture_bitrate_min: 32 + mp3_compression_on_mixture_bitrate_max: 320 + mp3_compression_on_mixture_backend: lameenc + all: + channel_shuffle: 0.5 + random_inverse: 0.01 + random_polarity: 0.5 + vocals: + pitch_shift: 1.0 + pitch_shift_min_semitones: -12 + pitch_shift_max_semitones: 12 + seven_band_parametric_eq: 0.5 + seven_band_parametric_eq_min_gain_db: -80 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.5 + tanh_distortion_min: 0.1 + tanh_distortion_max: 1 + time_stretch: 1.0 + time_stretch_min_rate: 0.5 + time_stretch_max_rate: 2 + bass: + pitch_shift: 1.0 + pitch_shift_min_semitones: -6 + pitch_shift_max_semitones: 6 + seven_band_parametric_eq: 0.4 + seven_band_parametric_eq_min_gain_db: -32 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 1.0 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + time_stretch: 1.0 + time_stretch_min_rate: 0.5 + time_stretch_max_rate: 1.5 + drums: + pitch_shift: 0.1 + pitch_shift_min_semitones: -6 + pitch_shift_max_semitones: 6 + seven_band_parametric_eq: 0.5 + seven_band_parametric_eq_min_gain_db: -24 + seven_band_parametric_eq_max_gain_db: 12 + tanh_distortion: 0.3 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + time_stretch: 1.0 + time_stretch_min_rate: 0.333 + time_stretch_max_rate: 1.5 + other: + pitch_shift: 1.0 + pitch_shift_min_semitones: -12 + pitch_shift_max_semitones: 12 + gaussian_noise: 0.4 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.15 + time_stretch: 0.01 + time_stretch_min_rate: 0.25 + time_stretch_max_rate: 1.5 +inference: + batch_size: 1 + dim_t: 256 + normalize: false + overlap_size: 33075 +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 147 + - 256 + - 512 + win_lengths: + - 1024 + - 2048 + - 4096 + window: hann_window + scale: mel + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 16.0 + w_log_mag: 16.0 + w_lin_mag: 16.0 + w_phs: 0.0 + mag_distance: L1 diff --git a/music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.ckpt b/music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6f85c5bb3908fa4a88351032885941e98355dbae --- /dev/null +++ b/music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca5f7c78543d84c369319f226b7337a6cc0b2e1a75d20409fc3ebde32686e94b +size 6411087858 diff --git a/music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.yaml b/music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..84a8b2b1a863aad5d7216b1fcf50fc4084fca5ec --- /dev/null +++ b/music_stems/four_stem/mel_band_roformer_4stems_xl_ver1.yaml @@ -0,0 +1,162 @@ +audio: + chunk_size: 661500 + dim_f: 1024 + dim_t: 1101 + hop_length: 882 + n_fft: 4096 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0001 +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 2049 + sample_rate: 44100 + stft_n_fft: 4096 + stft_hop_length: 882 + stft_win_length: 4096 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: false + skip_connection: true +training: + batch_size: 1 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + lr: 2.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 300 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + read_metadata_procs: 8 + normalize: false + other_fix: false + use_amp: true +augmentations: + enable: false + loudness: false + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true + mixup_probs: !!python/tuple + - 0.2 + - 0.02 + - 0.002 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + mp3_compression_on_mixture: 0.01 + mp3_compression_on_mixture_bitrate_min: 32 + mp3_compression_on_mixture_bitrate_max: 320 + mp3_compression_on_mixture_backend: lameenc + all: + channel_shuffle: 0.5 + random_inverse: 0.01 + random_polarity: 0.5 + vocals: + pitch_shift: 1.0 + pitch_shift_min_semitones: -12 + pitch_shift_max_semitones: 12 + seven_band_parametric_eq: 0.5 + seven_band_parametric_eq_min_gain_db: -80 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.5 + tanh_distortion_min: 0.1 + tanh_distortion_max: 1 + time_stretch: 1.0 + time_stretch_min_rate: 0.5 + time_stretch_max_rate: 2 + bass: + pitch_shift: 1.0 + pitch_shift_min_semitones: -6 + pitch_shift_max_semitones: 6 + seven_band_parametric_eq: 0.4 + seven_band_parametric_eq_min_gain_db: -32 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 1.0 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + time_stretch: 1.0 + time_stretch_min_rate: 0.5 + time_stretch_max_rate: 1.5 + drums: + pitch_shift: 0.1 + pitch_shift_min_semitones: -6 + pitch_shift_max_semitones: 6 + seven_band_parametric_eq: 0.5 + seven_band_parametric_eq_min_gain_db: -24 + seven_band_parametric_eq_max_gain_db: 12 + tanh_distortion: 0.3 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + time_stretch: 1.0 + time_stretch_min_rate: 0.333 + time_stretch_max_rate: 1.5 + other: + pitch_shift: 1.0 + pitch_shift_min_semitones: -12 + pitch_shift_max_semitones: 12 + gaussian_noise: 0.4 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.15 + time_stretch: 0.01 + time_stretch_min_rate: 0.25 + time_stretch_max_rate: 1.5 +inference: + batch_size: 1 + dim_t: 256 + normalize: false + overlap_size: 33075 +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 147 + - 256 + - 512 + win_lengths: + - 1024 + - 2048 + - 4096 + window: hann_window + scale: mel + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 16.0 + w_log_mag: 16.0 + w_lin_mag: 16.0 + w_phs: 0.0 + mag_distance: L1 diff --git a/music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.ckpt b/music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3ebffe428bb2daf33a76046feaaf946962f0bb4e --- /dev/null +++ b/music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e9daecd70aaed5b5a0d1f861cc4d77eaa45afb3fc6301b1cf32c1be0f5868fb +size 527385512 diff --git a/music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.yaml b/music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dd410adad5f922a74abe3ea22e0fb40921ca4118 --- /dev/null +++ b/music_stems/four_stem/model_bs_roformer_ep_17_sdr_9.6568.yaml @@ -0,0 +1,196 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 4 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 2 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.1 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.1 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.1 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.1 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 24255 diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt b/music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..45d7cc0edcc9c6e801ba4c2820054967d0fbbe9b --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3138fef1739d5a922d50258ea50805e5b55a3e795a5b593fcb0650980b5d8128 +size 250899372 diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.yaml b/music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c98efd9af72ab4164b5fe96c474e253df2a801b7 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_168_sdr_7.8127.yaml @@ -0,0 +1,152 @@ +audio: + chunk_size: 88200 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 6 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: True # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 16 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 4410 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt b/music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fd0681de0199b77c4ac6b04e94ca4ede4a8cb08e --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c50eb0498511fff2d7f3fc6d6385addae1d7cd6219c87a5be64f0a501e10eaf9 +size 234032732 diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.yaml b/music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7a53d50b9c694d5f159a40a1895c56291b9627a2 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_1_sdr_8.2175.yaml @@ -0,0 +1,106 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 4 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + +training: + batch_size: 5 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 17640 diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt b/music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..bbe8aef273345496f4aa201e895392af72f62593 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e00a25bbfb5caafa3a93c1d3133ad256e32e4021552c91e20d3804ca27a6f95 +size 81867982 diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.yaml b/music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.yaml new file mode 100644 index 0000000000000000000000000000000000000000..563676c558d0e5b0b9bb0098ce4daef707042a60 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_38_sdr_6.4698.yaml @@ -0,0 +1,150 @@ +audio: + chunk_size: 88200 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 128 + depth: 4 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + +training: + batch_size: 28 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 4410 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt b/music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..34b0dda50c64df565fce5ba7650417d8235117c4 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf2a6fd552005adc95b80f23d23517fa95a474c22f4490b31c4aafeee5003ca1 +size 81866140 diff --git a/music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.yaml b/music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8cb762310a4da2c8538a1334ae9ea9959a3a756c --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_ep_7_sdr_6.8652.yaml @@ -0,0 +1,152 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 128 + depth: 4 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 5 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 24255 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b95346d12997e7a7946606a8a1b79ed349078764 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7164a4962b6e43a53cec05dcc866428a77b111d1569b2ba7bba103bc55a0127c +size 81890712 diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.yaml b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2b65d67d14e476e0a9985e520dfe18a07fa9b1ef --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_15_sdr_6.4908.yaml @@ -0,0 +1,152 @@ +audio: + chunk_size: 176400 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 128 + depth: 4 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: True # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 8 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 8820 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b8f98c494d0f1ba3a142d13eb7a15754a0eba2e8 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8067adba1853a07cbd041c35915b08e9386fbd8516d619655a4ce458fef2636 +size 118636682 diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.yaml b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d60ae7935df5c9a90f4d90fd17acd62b45b6eeea --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_166_sdr_6.7022.yaml @@ -0,0 +1,152 @@ +audio: + chunk_size: 88200 + dim_f: 1024 + dim_t: 1101 + hop_length: 882 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 128 + depth: 4 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 2049 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 4096 + stft_hop_length: 882 + stft_win_length: 4096 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 20 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 4410 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..ef9cd1a9fe6466d4b068cc90d18c00f1e38fb244 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2faf9582ad8811ddc573925c54c91b5c8471cbe66805de404d5fb611d59adaac +size 118634392 diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.yaml b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.yaml new file mode 100644 index 0000000000000000000000000000000000000000..613781a1e18fbc3c124f04476111e58a4b180751 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_48_sdr_7.0412.yaml @@ -0,0 +1,152 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 1101 + hop_length: 882 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 128 + depth: 4 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 2049 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 4096 + stft_hop_length: 882 + stft_win_length: 4096 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 5 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 17640 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e74787cf660ed7522d990c77dfeeeb7bf7ed5d89 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6b2d654d257cb84d4546846253a098dc3cf17ce87568c00e35e41f48c9b52e7 +size 29889628 diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.yaml b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c555f6b617203305ea79920e5ccef1b0af9ce322 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_53_sdr_5.1235.yaml @@ -0,0 +1,152 @@ +audio: + chunk_size: 88200 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 64 + depth: 2 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 1 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 32 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 4410 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fb62585e1853ad523e0302e96692e4c189631ed9 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c25e58bca49df88438e96f9a7ac6f5e00cda64b7228384cd9af499ab9b88596 +size 177475814 diff --git a/music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.yaml b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.yaml new file mode 100644 index 0000000000000000000000000000000000000000..69231a0ece3ea024f62f43a61b5fa5953df45b10 --- /dev/null +++ b/music_stems/four_stem/model_mel_band_roformer_experimental_ep_9_sdr_6.5254.yaml @@ -0,0 +1,152 @@ +audio: + chunk_size: 176400 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 128 + depth: 4 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 2 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: True # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 8 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + normalize: false + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.01 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.01 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.01 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.01 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.01 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.01 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 8820 + normalize: false \ No newline at end of file diff --git a/music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.ckpt b/music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fce16a4c61735cd8b21d9943a000b4bbd1ab1235 --- /dev/null +++ b/music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac25975f0f5704f3d1a3c3c251505b7a0f417a22eafe82773440ee4f7e14b74f +size 214063778 diff --git a/music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.yaml b/music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b2953216f85f6750e7d33176f9ca5f4bc660785d --- /dev/null +++ b/music_stems/four_stem/model_scnet_ep_36_sdr_10.0891.yaml @@ -0,0 +1,185 @@ + +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: ['drums', 'bass', 'other', 'vocals'] + audio_channels: 2 + dims: [4, 64, 128, 256] + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: [0.230, 0.370, 0.400] + band_stride: [1, 4, 4] + band_kernel: [3, 4, 4] + conv_depths: [3, 3, 3] + compress: 4 + conv_kernel: 3 + num_dplayer: 8 + expand: 1 + +training: + batch_size: 3 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['drums', 'bass', 'other', 'vocals'] + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + # lr: 1.0 + optimizer: adam + lr: 5.0e-05 + normalize: false # perform normalization on input of model (use the same for inference!) + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + - 0.002 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + # apply mp3 compression to mixture only (emulate downloading mp3 from internet) + mp3_compression_on_mixture: 0.01 + mp3_compression_on_mixture_bitrate_min: 32 + mp3_compression_on_mixture_bitrate_max: 320 + mp3_compression_on_mixture_backend: "lameenc" + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + mp3_compression: 0.01 + mp3_compression_min_bitrate: 32 + mp3_compression_max_bitrate: 320 + mp3_compression_backend: "lameenc" + + # pedalboard reverb block + pedalboard_reverb: 0.01 + pedalboard_reverb_room_size_min: 0.1 + pedalboard_reverb_room_size_max: 0.9 + pedalboard_reverb_damping_min: 0.1 + pedalboard_reverb_damping_max: 0.9 + pedalboard_reverb_wet_level_min: 0.1 + pedalboard_reverb_wet_level_max: 0.9 + pedalboard_reverb_dry_level_min: 0.1 + pedalboard_reverb_dry_level_max: 0.9 + pedalboard_reverb_width_min: 0.9 + pedalboard_reverb_width_max: 1.0 + + # pedalboard chorus block + pedalboard_chorus: 0.01 + pedalboard_chorus_rate_hz_min: 1.0 + pedalboard_chorus_rate_hz_max: 7.0 + pedalboard_chorus_depth_min: 0.25 + pedalboard_chorus_depth_max: 0.95 + pedalboard_chorus_centre_delay_ms_min: 3 + pedalboard_chorus_centre_delay_ms_max: 10 + pedalboard_chorus_feedback_min: 0.0 + pedalboard_chorus_feedback_max: 0.5 + pedalboard_chorus_mix_min: 0.1 + pedalboard_chorus_mix_max: 0.9 + + # pedalboard phazer block + pedalboard_phazer: 0.01 + pedalboard_phazer_rate_hz_min: 1.0 + pedalboard_phazer_rate_hz_max: 10.0 + pedalboard_phazer_depth_min: 0.25 + pedalboard_phazer_depth_max: 0.95 + pedalboard_phazer_centre_frequency_hz_min: 200 + pedalboard_phazer_centre_frequency_hz_max: 12000 + pedalboard_phazer_feedback_min: 0.0 + pedalboard_phazer_feedback_max: 0.5 + pedalboard_phazer_mix_min: 0.1 + pedalboard_phazer_mix_max: 0.9 + + # pedalboard distortion block + pedalboard_distortion: 0.01 + pedalboard_distortion_drive_db_min: 1.0 + pedalboard_distortion_drive_db_max: 25.0 + + # pedalboard pitch shift block + pedalboard_pitch_shift: 0.01 + pedalboard_pitch_shift_semitones_min: -7 + pedalboard_pitch_shift_semitones_max: 7 + + # pedalboard resample block + pedalboard_resample: 0.01 + pedalboard_resample_target_sample_rate_min: 4000 + pedalboard_resample_target_sample_rate_max: 44100 + + # pedalboard bitcrash block + pedalboard_bitcrash: 0.01 + pedalboard_bitcrash_bit_depth_min: 4 + pedalboard_bitcrash_bit_depth_max: 16 + + # pedalboard mp3 compressor block + pedalboard_mp3_compressor: 0.01 + pedalboard_mp3_compressor_pedalboard_mp3_compressor_min: 0 + pedalboard_mp3_compressor_pedalboard_mp3_compressor_max: 9.999 + + vocals: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.1 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.2 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.33 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.33 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.1 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.1 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 363825 + normalize: false diff --git a/music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.ckpt b/music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e876a2ed0ea8262bd2bae422e8ab02b13558cba6 --- /dev/null +++ b/music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd889cc1d97619ccac59280ecc859c190cd3cc1b1557fbe3a19b1610bb67e410 +size 216189106 diff --git a/music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.yaml b/music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ec8612546b0b67c619ca09dc153902ef3fd98955 --- /dev/null +++ b/music_stems/four_stem/model_scnet_ep_54_sdr_9.8051.yaml @@ -0,0 +1,207 @@ +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - drums + - bass + - other + - vocals + audio_channels: 2 + dims: + - 4 + - 64 + - 128 + - 256 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.230 + - 0.370 + - 0.400 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 8 + expand: 1 + +training: + batch_size: 4 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + lr: 1.0e-05 + # lr: 1.0 + normalize: false # perform normalization on input of model (use the same for inference!) + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + # apply mp3 compression to mixture only (emulate downloading mp3 from internet) + mp3_compression_on_mixture: 0.01 + mp3_compression_on_mixture_bitrate_min: 32 + mp3_compression_on_mixture_bitrate_max: 320 + mp3_compression_on_mixture_backend: "lameenc" + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + mp3_compression: 0.01 + mp3_compression_min_bitrate: 32 + mp3_compression_max_bitrate: 320 + mp3_compression_backend: "lameenc" + + # pedalboard reverb block + pedalboard_reverb: 0.01 + pedalboard_reverb_room_size_min: 0.1 + pedalboard_reverb_room_size_max: 0.9 + pedalboard_reverb_damping_min: 0.1 + pedalboard_reverb_damping_max: 0.9 + pedalboard_reverb_wet_level_min: 0.1 + pedalboard_reverb_wet_level_max: 0.9 + pedalboard_reverb_dry_level_min: 0.1 + pedalboard_reverb_dry_level_max: 0.9 + pedalboard_reverb_width_min: 0.9 + pedalboard_reverb_width_max: 1.0 + + # pedalboard chorus block + pedalboard_chorus: 0.01 + pedalboard_chorus_rate_hz_min: 1.0 + pedalboard_chorus_rate_hz_max: 7.0 + pedalboard_chorus_depth_min: 0.25 + pedalboard_chorus_depth_max: 0.95 + pedalboard_chorus_centre_delay_ms_min: 3 + pedalboard_chorus_centre_delay_ms_max: 10 + pedalboard_chorus_feedback_min: 0.0 + pedalboard_chorus_feedback_max: 0.5 + pedalboard_chorus_mix_min: 0.1 + pedalboard_chorus_mix_max: 0.9 + + # pedalboard phazer block + pedalboard_phazer: 0.01 + pedalboard_phazer_rate_hz_min: 1.0 + pedalboard_phazer_rate_hz_max: 10.0 + pedalboard_phazer_depth_min: 0.25 + pedalboard_phazer_depth_max: 0.95 + pedalboard_phazer_centre_frequency_hz_min: 200 + pedalboard_phazer_centre_frequency_hz_max: 12000 + pedalboard_phazer_feedback_min: 0.0 + pedalboard_phazer_feedback_max: 0.5 + pedalboard_phazer_mix_min: 0.1 + pedalboard_phazer_mix_max: 0.9 + + # pedalboard distortion block + pedalboard_distortion: 0.01 + pedalboard_distortion_drive_db_min: 1.0 + pedalboard_distortion_drive_db_max: 25.0 + + # pedalboard pitch shift block + pedalboard_pitch_shift: 0.01 + pedalboard_pitch_shift_semitones_min: -7 + pedalboard_pitch_shift_semitones_max: 7 + + # pedalboard resample block + pedalboard_resample: 0.01 + pedalboard_resample_target_sample_rate_min: 4000 + pedalboard_resample_target_sample_rate_max: 44100 + + # pedalboard bitcrash block + pedalboard_bitcrash: 0.01 + pedalboard_bitcrash_bit_depth_min: 4 + pedalboard_bitcrash_bit_depth_max: 16 + + # pedalboard mp3 compressor block + pedalboard_mp3_compressor: 0.01 + pedalboard_mp3_compressor_pedalboard_mp3_compressor_min: 0 + pedalboard_mp3_compressor_pedalboard_mp3_compressor_max: 9.999 + + vocals: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.1 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.2 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.33 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.33 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.1 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.1 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 363825 + normalize: false diff --git a/music_stems/four_stem/model_scnet_sdr_9.3244.ckpt b/music_stems/four_stem/model_scnet_sdr_9.3244.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b72d6b1472f10f043fa266b4f3d6507ba55e4cc5 --- /dev/null +++ b/music_stems/four_stem/model_scnet_sdr_9.3244.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe550315a76e8f4aed8475d7d5952137504a3b6c63b3adcef2443bfe73aac540 +size 168868194 diff --git a/music_stems/four_stem/model_scnet_sdr_9.3244.yaml b/music_stems/four_stem/model_scnet_sdr_9.3244.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a2546f2ad5fa10548ade9cf683c09899f34afc87 --- /dev/null +++ b/music_stems/four_stem/model_scnet_sdr_9.3244.yaml @@ -0,0 +1,88 @@ +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - drums + - bass + - other + - vocals + audio_channels: 2 + dims: + - 4 + - 64 + - 128 + - 256 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.225 + - 0.372 + - 0.403 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 6 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + # lr: 1.0e-04 + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 363825 + normalize: false diff --git a/music_stems/four_stem/scnet_checkpoint_musdb18.ckpt b/music_stems/four_stem/scnet_checkpoint_musdb18.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..671b56f708c41055e2fd1ad71391254a8f097aac --- /dev/null +++ b/music_stems/four_stem/scnet_checkpoint_musdb18.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bc0d1abb20bfdf966dcd07637bafd03e4bc13653d09ef18bc9b3e342eafe2aa +size 42434986 diff --git a/music_stems/four_stem/scnet_checkpoint_musdb18.yaml b/music_stems/four_stem/scnet_checkpoint_musdb18.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e2bfdf8846c5dbd5df589da08165f2643d3eb7b6 --- /dev/null +++ b/music_stems/four_stem/scnet_checkpoint_musdb18.yaml @@ -0,0 +1,83 @@ +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - drums + - bass + - other + - vocals + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 10 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + lr: 5.0e-04 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 8 + dim_t: 256 + overlap_size: 363825 + normalize: true diff --git a/music_stems/four_stem/ver2.ckpt b/music_stems/four_stem/ver2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fe7a321b49416f94a730cd225cef4090e3ed2bb2 --- /dev/null +++ b/music_stems/four_stem/ver2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:835ce9fa6651769c28ad5a2c87f9220b003a461d7d6c57ebe0898fff5479138c +size 3764030874 diff --git a/music_stems/four_stem/ver2.yaml b/music_stems/four_stem/ver2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..650c002a96602d5df7dbcbe9c0d4710b23985fa2 --- /dev/null +++ b/music_stems/four_stem/ver2.yaml @@ -0,0 +1,70 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 882 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 4 + linear_transformer_depth: 0 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 2049 + sample_rate: 44100 + stft_n_fft: 4096 + stft_hop_length: 882 + stft_win_length: 4096 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: false + skip_connection: true +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - other + - vocals + lr: 1.0 + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + read_metadata_procs: 8 + normalize: false + other_fix: false + use_amp: true +inference: + batch_size: 4 + dim_t: 1101 + normalize: false + overlap_size: 24255 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2865a11fe4a9b38672aa29c03e64c663449cd216 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aafcbe186aaec0489d28e388d9d1d3c12660c5640a8494bdaf3cbfee60c08c78 +size 77619808 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b13a70ca54c600e57ab952b9bd6340cb841c6135 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_accordion_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - accordion + - other + target_instrument: accordion + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d6e0677ba8ef8772a76eed45b81e01fea09b2266 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa386b2e7b1ea4f12b9b5c557444c0dc78648ef4ee299de2759d86457e182b3e +size 77624038 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..110840eab9486a992730965c75fc104e9c54893d --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_acoustic-guitar_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - acoustic-guitar + - other + target_instrument: acoustic-guitar + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f668a047da7251446d3b4c5fd007f0770d4c61e8 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e60a9057340574977b0f858084113d952ba4731a7bbe054bf49d1d5338ca9d9 +size 77620513 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..758e35ee997a7bd51c5f1134e1497d63f538fa9c --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_back-vocal_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - back-vocal + - other + target_instrument: back-vocal + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c7817bccab8b01e0443b9a37f22bc37dc77eb2c1 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f9d0ad56db9289c767d41104156c2dce5b88631e05958bc78479b77e62e4f0d +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..43d1ef411870bfb8184271a067efb96554dbe6ef --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_banjo_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - banjo + - other + target_instrument: banjo + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..7d09116a33c34cabb71c4eec02cedf5c1ab3993c --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b17600dd1c2027a72ab34180ff6191d3bdeb786db9fbb13c58337c33fc10163c +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..250fe6d85cc914e8d59b3f0cbfeb3b1d822d245e --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bass_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - bass + - other + target_instrument: bass + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..41f8d27f15e722b21aa681fcf97c722ff9a03c51 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d7329356a80027377e11d6d20c1c37a6b1c8a97e6fc502379142ba88773caae +size 77618398 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..314d8c3f1ff9c23dcfb3352cb43feb3bbcd13ef7 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bassoon_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - bassoon + - other + target_instrument: bassoon + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c9c88f3f6b187179aba99724d100703b62bf43b2 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6328f387259bb50ef12e36d3ef6cc5703520a0a89cd3bd743cf4f420f87701b8 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..de42d743d11833b163340edd80fac33d3eaf4464 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bells_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - bells + - other + target_instrument: bells + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..960a338ca486f758f7fd68f32c0028e856f43e59 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:598a644d78dd6a385140b2134733eef42d9525805a84d55f224c80a1ca39ff61 +size 77622628 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..31ae25d4bd91168c54b9d9a195b32341c7a8a29d --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_bowed_strings_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - bowed_strings + - other + target_instrument: bowed_strings + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1e0cf7f648b917e8587a4f6675566680dcf48307 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7d7bf86a031f3253019ff9dc424485d8085946fc428ab66522c568ee17b8027 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d6dec1c272e6dacda317b608f9738403b9c093b2 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_brass_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - brass + - other + target_instrument: brass + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..375ff54f14e7938eb926d7386d6d863e99a5fcdf --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb6600ebc4dfe1d896d221073d114f51902a0b385afa23503034b9e8befba62c +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9b66d8a575b5d0a3071d5dd14faac30941fa2ba0 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_cello_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - cello + - other + target_instrument: cello + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..bd090433d0f8516aebd6ae2c99e63b60f6644dff --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c4ac033c03727943a491e05dc9c9aebbd92afc3d8e83b3484881e9ed3f3dde7 +size 77619103 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0d2c3a7680acfee32686790e336a9efa20608398 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_clarinet_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - clarinet + - other + target_instrument: clarinet + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..cc7914f21aa1a408d31a7e8ed94ed29555ffa231 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78885092a3fdf0f63f349ba34e8fa4ff1ef84959a7e94f544b437e2404af5287 +size 77617693 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..abf16779c700bf13a71cda61f6e5ba19ade15ed7 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_congas_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - congas + - other + target_instrument: congas + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..bf051f5669a947c9d59596bb4a7d592ecd3f532b --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:481c741ff36208c9b11860297ff5c9a502873662724d35a437bf64aaade890bd +size 77622628 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9a7de291181c68963103bb15204e5f60b6208ade --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_digital-piano_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - digital-piano + - other + target_instrument: digital-piano + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..eb07649a950cf3ed19f7c75fbc5aa45fbe8acc69 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8b99b1b837f04200cbb5b99d33c0ea2ac46bb76424879385d8fc6dde8abfe0b +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9c33cbf111243768e993613a00f6058475929d92 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_dobro_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - dobro + - other + target_instrument: dobro + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e153fcdf6a80d0767d790b1e5639e5c1985998e9 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:731d7bcaae4ce02f5a19116c1b5c1004bfe7208d14cc6b93aaab2753c0302cfb +size 77621218 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..90e79b3df28aafa6b0bbade188c2dd2b5b36d397 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_double-bass_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - double-bass + - other + target_instrument: double-bass + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..634f3c47575bf393713848ac92b8103740e514a6 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abfa64203417f0874aa3936320b1666364ef1badd917f5613a3017d0ea73c7a2 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f0e0471681fcaa4f889741fdca076486d1623251 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_drums_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - drums + - other + target_instrument: drums + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..15780b326cdad5a1fd589f46185d25b018748c23 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd506bfce9474f91a31001967f2c4935ce4e67f643da3df20d04058da927c553 +size 77624038 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8540bd8cb0321de4f8748e69bfbbf96ac9ad3788 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_electric-guitar_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - electric-guitar + - other + target_instrument: electric-guitar + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a3ce92fe7d317571ed840aa3f5a41642d92d9b50 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65dfa1e2c093165812cbdf8945715e7f71566bc3a8196ecad3edce1d29bb2e45 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..854b337477a3c69630316a6a2d75581ab40eb0da --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_flute_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - flute + - other + target_instrument: flute + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c323b8a60032312b41a89b612bfb2243fda6cd57 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:484b9e68a962891daffb1d671f847509a7e6c7f72de42c445c774881dfe2c420 +size 77621218 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..aaa61ff9029f1190b81aaf7895fe68d3fbf24cc1 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_french-horn_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - french-horn + - other + target_instrument: french-horn + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d4e49448a4bdc84157af1ef17b4d4b6f7f5e8d0e --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc18ac41fb8e0e1cf2dcb290d67c9e2f7a295f380e91864eda9f3cc805015231 +size 77621923 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..15217ec7eb6c45d93f139a978d40b054b7ada0e3 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_glockenspiel_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - glockenspiel + - other + target_instrument: glockenspiel + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3bc878cdb67695f0de1b0d79201387b7b996f42e --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e4524d60cc7eb01eae4048ecda6e3fbf9a9a404e12838f4ed58b8dfb733762d +size 77617693 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bd5cc0580d45bdc7566c5207791f9667a8c96a5c --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_guitar_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - guitar + - other + target_instrument: guitar + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..bf62f9c49c3720cbfbc6839c1f0462738df0eabe --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90f9e794573c6492695efb18c5bd8967f28f7a1832aa9d237a2ebe53ffd7347c +size 77619808 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2c9774afdbf5cc1fe1ee4130b02febf0c35e04e2 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_harmonica_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - harmonica + - other + target_instrument: harmonica + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c87de7830454cb96c46a96d9688c0f73e00129e5 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fde42f720b062391e5a62df3b0ab86244bb0a6471b4896db3677e88cbf90ef6 +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6d556f3667332704e6fccc247a7deb6891eaf354 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_harp_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - harp + - other + target_instrument: harp + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..11b337589cfd6187e920b5432e73ec48657f9b04 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:362f6c681a465aea21d3d4b8d8eba29e0e2ae41152bf0b8798201cfb00f32ee1 +size 77621218 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c2dda291b583a59f2db897f381ef0776a34901c4 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_harpsichord_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - harpsichord + - other + target_instrument: harpsichord + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3aa6be53a595e7771ac533f8c2709e83d7f82365 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94901a05d358f16b2de4064eec45f0947bb28bf239040a3fe79e2ae7fadd2984 +size 77614809 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3da0967eadba72dc02c3a82531a700f297b2f382 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_hh_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - hh + - other + target_instrument: hh + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1a87bbb5056722a0422cd9d83eff60b2b51ace5c --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bc2bde8c5b26d5e22720304deace45dfeaffb2d02dc07922a8f2be17d82c844 +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d1bdbabea95f48d70cf329a786c6452d610ae59c --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_keys_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - keys + - other + target_instrument: keys + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6b2e967717143a69780cc9603bf4ca45c3432565 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad914b83f9ef13bedf9608771f4f1a86174e27a84f7f235dc2d1dd91bab85fbc +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bec3978e689878b1f95267b6074135dcab304d06 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_kick_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - kick + - other + target_instrument: kick + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..cc49f035e289b5c0ebad7c311be1b0675c101b22 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7622ef83040f3a1e683d4250af10bcc52ebc36f0b4188ff2101d23c6dae310ce +size 77620513 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cec0cefba6dceb2b018bf35cc391eec97c5ee129 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_lead-vocal_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - lead-vocal + - other + target_instrument: lead-vocal + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a199c0e5c4e1db85a21be66cfb67365f5f753177 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a5589c6ee51a8f3f87f68e48aaf41792c31f63d5c0eaba497ae0ecd4bb6a675 +size 77619103 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fb1d1bac98bde0905dcf11a0604ba309e8ae3d41 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_mandolin_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - mandolin + - other + target_instrument: mandolin + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4c27a25ae0a479f8e1b45d61ff62ea9d4430ecc6 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:559835d6525898228f1ebcfac770ed44185217e547159a3c0f836391637fc3e8 +size 77618398 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..78fac03ba389b2d5ebfc6d60f4a0d495236f067a --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_marimba_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - marimba + - other + target_instrument: marimba + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9a6e187de134e24a2f5141eb2e3e6a86590fef85 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7de88790f82b86af910f943f5a31f30237c48997011e6123c5922449f2a22285 +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..28eff90794c85b6a45323e7da625e1ee8f8a0b40 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_oboe_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - oboe + - other + target_instrument: oboe + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e23ec8ed1c4b6ed6181cb39beb925eaba7256247 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:14c1c2a26fbd5061be85a31a170b6989ba296871895905ea6a1a048d200895ba +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d1a2a4309c897e6f5c38281d3e2fec569845ce1d --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_organ_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - organ + - other + target_instrument: organ + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4657fb1085ff054252bb2c5b8a2d6e3b7dff28e4 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82187665185d51e6050b4d6afaf45e7acafd85d1eefffd86c47ac0aa601e089f +size 77620513 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..05de69cd2aa52b90efa13f08c1d59a34e07590b6 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_percussion_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - percussion + - other + target_instrument: percussion + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..96bf416f526304318c1dd71ab5d6f5c424b705fe --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e58ad544386136028ec1f5764bb95a235235492144112d94782ff1333905fe9 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..19bcd39d1f3bebb78d37f14ded9698faff078c48 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_piano_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - piano + - other + target_instrument: piano + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9cdeb1507b2d7f9e3b03fe2ea78a23d69cda04ef --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7027da3852828194f224f5f822582a38015b0afc452fe3e407229084c10de6a9 +size 77619808 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..053af765035d189901b2486636cd1b2391dc4621 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_saxophone_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - saxophone + - other + target_instrument: saxophone + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1586ed689349935c49048bdc958a1418904a56d0 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32683e3fcddebe1a5605fb5cffc76f8e89ec19d35f0ed49fe6f0f31476fd37d1 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fc072f3412155a60b3c1a1b505b3ff6dc6d3799e --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_sitar_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - sitar + - other + target_instrument: sitar + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..dac516a05cca02f5276700d857b3ad045e45ba78 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4cde14867f1311836d4aa07649e39df4df43c4ba98780783fe96cd4fe8293bf +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..28810be99c0614e0404cf01a9638e4834005cc51 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_snare_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - snare + - other + target_instrument: snare + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6d4551e975848e89323d5519420c029ec93f9768 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e8b76ff287f6127e1f3dd927b56946784d368d5a40e7f4f74b1373a647675c5 +size 77618398 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ef0bc3120cb34d287f1417c964e7a8ee7f8a476d --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_strings_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - strings + - other + target_instrument: strings + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..46adb36acb7fbdd544f8190203db2123c1b49374 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04749df7e2fab770a0d26bf9aa4850c1a6cd89547fc8bfe9c9d223602c0afb82 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bc5792a547f5ac9973cb628d0e3d5b50e2ef8f2b --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_synth_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - synth + - other + target_instrument: synth + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..49d087ba3be8bc44ea586a4315648b103cee1a83 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e22d5e185ab3db1262f1fc6b98eff22d9f97fc6f6e7f381cacf4624e6b69ad2 +size 77620513 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..41ad6e914c781b2786467f3528e38ebaec3c2567 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_tambourine_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - tambourine + - other + target_instrument: tambourine + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..65db829a870f0c0f6a5866b45720ccc621f4fb3f --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2f4813c366246f3af010340d9f1c7854bcb48e3187e51065dfaff34ca515b54 +size 77618398 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0e09b2bdf9fdda116a6285830137a37945a0c478 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_timpani_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - timpani + - other + target_instrument: timpani + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c58fda03fc3de8267d18ed1e04915d3ac225e0bc --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ceb45bbc16419363bb842484dadcc786a9ca290c4553c9eced0e98dc8f0f2458 +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..49f9fbe9d07c3349adc9dbda0291e6bbdedb9df6 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_toms_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - toms + - other + target_instrument: toms + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..13a6ff382e16ffa82864104551239fd1a8eb2945 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e3692157868c6c7d0d4d6f36be263ed363e01cf66e4c97a400acb8c05ca250f +size 77619103 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a97d6296d765646caf3a7db3073a4ab2aef24c49 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_triangle_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - triangle + - other + target_instrument: triangle + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4bc29606fdc8c8aed30d0b5c532638c4a4aa68c7 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f50d2e13e3e02cd74c3a5e084975d64a2a1f2090757c64495944a00823947f0d +size 77619103 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..72bab37592c13d0ef42886f8dcd3ec6ec42d6fd1 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_trombone_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - trombone + - other + target_instrument: trombone + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..64b8fb504c16f107a5b7b9bcc16a20114b5a4601 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47530e8c6ee14eb3f4469efc629b0a3942cfff062f3a7eb788cb15355fd7b724 +size 77618398 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0c64bc2f9ddc1cc48b5e4a0d2ac1f52abd1f8d43 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_trumpet_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - trumpet + - other + target_instrument: trumpet + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..df8cc7df7c7c5746c65e59c2fb8fc3b207b6c164 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0bb341f247101ba9e723a57089e0bdade930eab9fa09c349d1d36c3272bce34f +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..596ef17dbd23192b4593db5adfe3b41a09463fd9 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_tuba_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - tuba + - other + target_instrument: tuba + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4bfff155bc06e21c3e598b74f1c64fa6b1eb5273 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56eb5a2cb4c3a45be630c27b03815ead66e9a892922b84f9ac27eeb1fe0be2f0 +size 77618398 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..88e347b36f9bce0ff9fb46414503c3b0aa064f3a --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_ukulele_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - ukulele + - other + target_instrument: ukulele + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..884d814b839f04b992afcdfd26f2c05a28f32c3a --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41d47d879a45fe94bba0db01476cce5e93395b06c756b98463b1eda95a468354 +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..80219fea1ff04919692681b0a7dfd7cb0e1c263d --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_viola_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - viola + - other + target_instrument: viola + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..28ad85b8762aaf6d01dc264f6ad8e4a94491552f --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43dd11a1816ba7c58d74f4de1a48ffc05b98fc499dfb24cf8e71c5ba729f2ada +size 77617693 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3f1cc3c5d4895f5bbd560b2098f15f7ee7e23e6c --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_violin_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - violin + - other + target_instrument: violin + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..357f16792a395db8dae3167bcaa753b0d1f4988d --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:917580ad6a1bab75d1b438db537ab98387839efb1809e5a50c78b7afdb96c8af +size 77616988 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e50a447d6911d536409ce283b6c3372290ec4a89 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_vocal_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - vocal + - other + target_instrument: vocal + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f720562c0fdaa45269260e37688bce470aa9a636 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9bcb89de76916a44243e89ffd02051b527848afd8d2d618bd82e04de3bd861f3 +size 77621218 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..132160b13046b806d8d94c3e6b15f59177ad0252 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_wind-chimes_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - wind-chimes + - other + target_instrument: wind-chimes + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f7922903936d2227c1ffdc07df8546bbbaed91a7 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2ae2f8b7aa777dba3596fa8a8c0aec42b1bf3c493ebdc9aaf2daeffcab9d025 +size 77616219 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a8de6a7beab2684e9ecf58456ceb1f8743c5a7a1 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_wind_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - wind + - other + target_instrument: wind + use_amp: true diff --git a/music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.ckpt b/music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..44fd02568124a6172042f77fe35ffb203094e6c3 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a1335d2cd21cd4fdd6964e7ab96c9e9776b2643644f763e2f59bb6c53d39060 +size 77619103 diff --git a/music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.yaml b/music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.yaml new file mode 100644 index 0000000000000000000000000000000000000000..62652b354af39c3b36cd3cbb8039de26f6d81b83 --- /dev/null +++ b/music_stems/mega_53_stems/bs_mega_53stem_woodwind_mvsep.yaml @@ -0,0 +1,116 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + chunk_size: 882000 + dim_t: 1101 + normalize: false + overlap_size: 22050 +model: + attn_dropout: 0.0 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.0 + flash_attn: true + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + mlp_expansion_factor: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_stems: 1 + skip_connection: false + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 + use_torch_checkpoint: false +training: + batch_size: 1 + instruments: + - woodwind + - other + target_instrument: woodwind + use_amp: true diff --git a/music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt b/music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fb68ba0585ede03491bc33ceb8c76eedc970cd8f --- /dev/null +++ b/music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c62820893bbf86d4e734f966bd142d9157cfc8bb8e79e9d8f9ea553f3ff3519f +size 1368919887 diff --git a/music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.yaml b/music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..adff3b65ba2c83eb1c92db7e5db894bc9858405e --- /dev/null +++ b/music_stems/mega_53_stems/mvsep_mega_model_bs_roformer_53_stems_v1.yaml @@ -0,0 +1,231 @@ +audio: + chunk_size: 441000 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 53 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 2 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + model_type: 'bs_roformer' + batch_size: 1 + gradient_accumulation_steps: 2 + grad_clip: 0 + instruments: [ + 'accordion', + 'acoustic-guitar', + 'back-vocal', + 'banjo', + 'bass', + 'bassoon', + 'bells', + 'bowed_strings', + 'brass', + 'cello', + 'clarinet', + 'congas', + 'digital-piano', + 'dobro', + 'double-bass', + 'drums', + 'electric-guitar', + 'flute', + 'french-horn', + 'glockenspiel', + 'guitar', + 'harmonica', + 'harp', + 'harpsichord', + 'hh', + 'keys', + 'kick', + 'lead-vocal', + 'mandolin', + 'marimba', + 'oboe', + 'organ', + 'percussion', + 'piano', + 'saxophone', + 'sitar', + 'snare', + 'strings', + 'synth', + 'tambourine', + 'timpani', + 'toms', + 'triangle', + 'trombone', + 'trumpet', + 'tuba', + 'ukulele', + 'viola', + 'violin', + 'vocal', + 'wind', + 'wind-chimes', + 'woodwind' + ] + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 240 + num_steps: 6000 + # precompute_chunks: True + num_precompute_chunks: 100000 + q: 0.95 + coarse_loss_clip: true + optimizer: prodigy + lr: 1.0 + # optimizer: adam + # lr: 0.9e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.4 + loudness_max: 1.6 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + - 0.002 + - 0.0002 + mixup_loudness_min: 0.4 + mixup_loudness_max: 1.6 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.01 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 147 + - 256 + - 512 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: false + w_sc: 0.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 1.0 + mag_distance: "L1" + + +inference: + chunk_size: 882000 + batch_size: 1 + dim_t: 1101 + overlap_size: 22050 + normalize: false diff --git a/music_stems/six_stem/BS-Roformer-SW.ckpt b/music_stems/six_stem/BS-Roformer-SW.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..011736ea342314bf6193d061f60929f95f4f66bd --- /dev/null +++ b/music_stems/six_stem/BS-Roformer-SW.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24e7d35ee9c64415673d3fd33e06a67cac2c103c5df6267ba1576459c775916e +size 699412152 diff --git a/music_stems/six_stem/BS-Roformer-SW.yaml b/music_stems/six_stem/BS-Roformer-SW.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7af9637d000b8f34c7ae4f29fe8c94f1ae5fb971 --- /dev/null +++ b/music_stems/six_stem/BS-Roformer-SW.yaml @@ -0,0 +1,198 @@ +audio: + chunk_size: 588800 #882000 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 6 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['bass', 'drums', 'other', 'vocals', 'guitar', 'piano'] + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.1 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.1 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.1 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.1 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +inference: + batch_size: 1 + dim_t: 801 # Changed from 1101 to match training + overlap_size: 29440 + normalize: false + diff --git a/music_stems/six_stem/HTDemucs4_6stems.th b/music_stems/six_stem/HTDemucs4_6stems.th new file mode 100644 index 0000000000000000000000000000000000000000..43e9de1c47734b05e2e2d9e8f2a70a548970326a --- /dev/null +++ b/music_stems/six_stem/HTDemucs4_6stems.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:34c22ccb381c6f9fdbf324f04e1e2fe21aaaf293f5ded163a162697ff9a02ddd +size 54996327 diff --git a/music_stems/six_stem/HTDemucs4_6stems.yaml b/music_stems/six_stem/HTDemucs4_6stems.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7520ce5184f4fbdfa8e905cc08bc66bf6cf4bd35 --- /dev/null +++ b/music_stems/six_stem/HTDemucs4_6stems.yaml @@ -0,0 +1,114 @@ +audio: + chunk_size: 485100 + hop_length: 1024 + min_mean_abs: 0.001 +augmentations: + all: + channel_shuffle: 0.5 + random_inverse: 0.1 + random_polarity: 0.5 + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 + mixup: true + mixup_loudness_max: 1.5 + mixup_loudness_min: 0.5 + mixup_probs: + - 0.2 + - 0.02 +htdemucs: + bottom_channels: 0 + cac: true + channels: 48 + channels_time: null + context: 1 + context_enc: 0 + dconv_comp: 8 + dconv_depth: 2 + dconv_init: 1e-3 + dconv_mode: 3 + depth: 4 + emb_scale: 10 + emb_smooth: true + end_iters: 0 + freq_emb: 0.2 + growth: 2 + kernel_size: 8 + multi_freqs: [] + multi_freqs_depth: 3 + nfft: 4096 + norm_groups: 4 + norm_starts: 4 + num_subbands: 1 + rescale: 0.1 + rewrite: true + stride: 4 + t_auto_sparsity: false + t_cape_augment: true + t_cape_glob_loc_scale: + - 5000.0 + - 1.0 + - 1.4 + t_cape_mean_normalize: true + t_cross_first: false + t_dropout: 0.0 + t_emb: sin + t_gelu: true + t_global_window: 100 + t_group_norm: false + t_heads: 8 + t_hidden_scale: 4.0 + t_layer_scale: true + t_layers: 5 + t_lr: null + t_mask_random_seed: 42 + t_mask_type: diag + t_max_period: 10000.0 + t_max_positions: 10000 + t_norm_first: true + t_norm_in: true + t_norm_in_group: false + t_norm_out: true + t_sin_random_shift: 0 + t_sparse_attn_window: 400 + t_sparse_cross_attn: false + t_sparse_self_attn: false + t_sparsity: 0.95 + t_weight_decay: 0.0 + t_weight_pos_embed: 1.0 + time_stride: 2 + wiener_iters: 0 + wiener_residual: false +inference: + batch_size: 1 + overlap_size: 363825 +model: htdemucs +training: + batch_size: 8 + channels: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - drums + - bass + - other + - vocals + - guitar + - piano + lr: 9.0e-05 + normalize: true + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + samplerate: 44100 + segment: 11 + shift: 1 + target_instrument: null + use_amp: true diff --git a/music_stems/six_stem/logic_bs_roformer.ckpt b/music_stems/six_stem/logic_bs_roformer.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8dba48e1a980882cd0784355fe07a932b45c8143 --- /dev/null +++ b/music_stems/six_stem/logic_bs_roformer.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:57c4a987b7e99b4d313018eabf7a8b7b585bfbb64e809657a7e94faede62333a +size 699446752 diff --git a/music_stems/six_stem/logic_bs_roformer.yaml b/music_stems/six_stem/logic_bs_roformer.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e68700e9dbd14bb9898f7490ac1c2f8172074ca8 --- /dev/null +++ b/music_stems/six_stem/logic_bs_roformer.yaml @@ -0,0 +1,197 @@ +audio: + chunk_size: 588800 #882000 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 6 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + use_shared_bias: True +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['bass', 'drums', 'other', 'vocals', 'guitar', 'piano'] + patience: 3 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + vocals: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + bass: + pitch_shift: 0.1 + pitch_shift_min_semitones: -2 + pitch_shift_max_semitones: 2 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -3 + seven_band_parametric_eq_max_gain_db: 6 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.5 + drums: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.1 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.6 + other: + pitch_shift: 0.1 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.1 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.1 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 29440 + normalize: false \ No newline at end of file diff --git a/reverb_echo_control/dereverb/Lead_VocalDereverb.ckpt b/reverb_echo_control/dereverb/Lead_VocalDereverb.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..7746ea45ad49454bbcbf2e2a57a7f1d2b5cadeaf --- /dev/null +++ b/reverb_echo_control/dereverb/Lead_VocalDereverb.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:097c13b298414905651867ef315d881bbf2604a5c04be28820ed5946338b3eb4 +size 913031195 diff --git a/reverb_echo_control/dereverb/Lead_VocalDereverb.yaml b/reverb_echo_control/dereverb/Lead_VocalDereverb.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7dc4a82a4149bb36f78b0dac076bef471091e800 --- /dev/null +++ b/reverb_echo_control/dereverb/Lead_VocalDereverb.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 3 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 4000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: true + use_amp: true +augmentations: + enable: true + loudness: true + loudness_min: 0.1 + loudness_max: 1.0 + mixup: false + mixup_probs: !!python/tuple + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/reverb_echo_control/dereverb/MDX23C-De-Reverb-aufr33-jarredou.ckpt b/reverb_echo_control/dereverb/MDX23C-De-Reverb-aufr33-jarredou.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..dca4c63709a57308193ba52a2d3309efed5915bc --- /dev/null +++ b/reverb_echo_control/dereverb/MDX23C-De-Reverb-aufr33-jarredou.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eae2471b707758d74db38ac1b1d5800e12f57c4e9d1ebbb2faf004b8e086e914 +size 448098867 diff --git a/reverb_echo_control/dereverb/MDX23C-De-Reverb-aufr33-jarredou.yaml b/reverb_echo_control/dereverb/MDX23C-De-Reverb-aufr33-jarredou.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ffeb02359800372de9ce6c1b14a32cd625c5013b --- /dev/null +++ b/reverb_echo_control/dereverb/MDX23C-De-Reverb-aufr33-jarredou.yaml @@ -0,0 +1,136 @@ +audio: + chunk_size: 261120 + dim_f: 4096 + dim_t: 256 + hop_length: 1024 + n_fft: 8192 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + act: gelu + bottleneck_factor: 4 + growth: 128 + norm: InstanceNorm + num_blocks_per_scale: 2 + num_channels: 128 + num_scales: 5 + num_subbands: 4 + scale: + - 2 + - 2 + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - dry + - No dry + lr: 1.0e-06 + patience: 4 + reduce_factor: 0.93 + target_instrument: null + num_epochs: 40 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adamw + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.05 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + + # pedalboard chorus block + pedalboard_chorus: 0.001 + pedalboard_chorus_rate_hz_min: 1.0 + pedalboard_chorus_rate_hz_max: 7.0 + pedalboard_chorus_depth_min: 0.25 + pedalboard_chorus_depth_max: 0.95 + pedalboard_chorus_centre_delay_ms_min: 3 + pedalboard_chorus_centre_delay_ms_max: 10 + pedalboard_chorus_feedback_min: 0.0 + pedalboard_chorus_feedback_max: 0.01 + pedalboard_chorus_mix_min: 0.1 + pedalboard_chorus_mix_max: 0.9 + + # pedalboard phazer block + pedalboard_phazer: 0.001 + pedalboard_phazer_rate_hz_min: 1.0 + pedalboard_phazer_rate_hz_max: 10.0 + pedalboard_phazer_depth_min: 0.25 + pedalboard_phazer_depth_max: 0.95 + pedalboard_phazer_centre_frequency_hz_min: 200 + pedalboard_phazer_centre_frequency_hz_max: 12000 + pedalboard_phazer_feedback_min: 0.0 + pedalboard_phazer_feedback_max: 0.5 + pedalboard_phazer_mix_min: 0.1 + pedalboard_phazer_mix_max: 0.9 + + # pedalboard pitch shift block + pedalboard_pitch_shift: 0.01 + pedalboard_pitch_shift_semitones_min: -7 + pedalboard_pitch_shift_semitones_max: 7 + + # pedalboard resample block + pedalboard_resample: 0.001 + pedalboard_resample_target_sample_rate_min: 4000 + pedalboard_resample_target_sample_rate_max: 44100 + + mp3_compression_min_bitrate: 32 + mp3_compression_max_bitrate: 320 + mp3_compression_backend: "lameenc" + + dry: + # pedalboard distortion block + pedalboard_distortion: 0.001 + pedalboard_distortion_drive_db_min: 1.0 + pedalboard_distortion_drive_db_max: 25.0 + + tanh_distortion: 0.05 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + # pedalboard bitcrash block + pedalboard_bitcrash: 0.005 + pedalboard_bitcrash_bit_depth_min: 4 + pedalboard_bitcrash_bit_depth_max: 16 + + seven_band_parametric_eq: 0.24 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + + gaussian_noise: 0.005 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.01 + + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + other: + seven_band_parametric_eq: 0.24 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + +inference: + batch_size: 2 + dim_t: 256 + num_overlap: 4 + overlap_size: 195840 \ No newline at end of file diff --git a/reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.ckpt b/reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..814f35f8718aab66733f43e60a034fa933dd2f7f --- /dev/null +++ b/reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0506455e74ffc02bbec700df9863ae243597034003815f1418227c6dee33b6ea +size 455864012 diff --git a/reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.yaml b/reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12a11d6b15a59acc5e99f68e39ca5f40ff4731f2 --- /dev/null +++ b/reverb_echo_control/dereverb/de_big_reverb_mbr_ep_362.yaml @@ -0,0 +1,62 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 +model: + attn_dropout: 0.1 + depth: 8 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.1 + flash_attn: true + freq_transformer_depth: 1 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_bands: 60 + num_stems: 1 + sample_rate: 44100 + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + batch_size: 1 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 8 + instruments: + - dry + - other + lr: 4.0e-05 + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: dry + use_amp: true diff --git a/reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt b/reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5431ef75c131a758c45710f4a14763c5b0a5b648 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4dfa4c4aa251c0afc303e5d512d904a1c2255e51d8d0dbfbddb18d740563d42 +size 204496717 diff --git a/reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.yaml b/reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.yaml new file mode 100644 index 0000000000000000000000000000000000000000..125a24210cc4cafab69fae20da083a93815574e0 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_bs_roformer_anvuew_sdr_22.5050.yaml @@ -0,0 +1,126 @@ +audio: + chunk_size: 448000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + use_torch_checkpoint: false + skip_connection: false +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 1 + reduce_factor: 0.75 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false + use_amp: true +inference: + batch_size: 2 + dim_t: 876 + overlap_size: 22400 diff --git a/reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8f21995361b651a6c446cb4e1e7664d94eb42bff --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9262877b87e9ebb0fb808a456b0a411fa677f5df31c8383c1254af531c078970 +size 913107578 diff --git a/reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.yaml b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0fb9e4284701f8372479e45663b90035a09eeb95 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_anvuew_sdr_19.1729.yaml @@ -0,0 +1,76 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 3 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 4000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.1 + loudness_max: 1.0 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..45d5062aae81caa8df3565e7d10c818f076cc588 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0db8f1b41c00cead1112e967262a12802fd32e76c0c3a8eb207e772bae25d07b +size 913107578 diff --git a/reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.yaml b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0fb9e4284701f8372479e45663b90035a09eeb95 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_less_aggressive_anvuew_sdr_18.8050.yaml @@ -0,0 +1,76 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 3 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 4000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.1 + loudness_max: 1.0 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.ckpt b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..06855114df2802baa49c15faf4b39a743b84e27b --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f099ee717eb57fb0ad5eb0e7c9ad6787c36168140b61ce2b158b90c2c4ecee79 +size 913097978 diff --git a/reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.yaml b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0fb9e4284701f8372479e45663b90035a09eeb95 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_mel_band_roformer_mono_anvuew.yaml @@ -0,0 +1,76 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 3 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 4000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.1 + loudness_max: 1.0 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.ckpt b/reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6157fee3cb7108052c09f8fe505f034779c698d4 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2edec521f09e26341c1923dc82c8c52dbc86478b42b9999f679535743c970cb3 +size 118128452 diff --git a/reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.yaml b/reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f1d9ffe02bbc43a8e42d682d792395d97e171038 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_room_anvuew_sdr_13.7432.yaml @@ -0,0 +1,132 @@ +audio: + chunk_size: 384000 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 1 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 128 + depth: 16 + stereo: false + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 3 + - 3 + - 3 + - 3 + - 3 + - 4 + - 4 + - 4 + - 4 + - 4 + - 5 + - 5 + - 5 + - 5 + - 6 + - 6 + - 6 + - 6 + - 7 + - 7 + - 7 + - 8 + - 8 + - 8 + - 9 + - 9 + - 10 + - 10 + - 11 + - 12 + - 13 + - 14 + - 15 + - 16 + - 17 + - 18 + - 19 + - 20 + - 21 + - 22 + - 23 + - 24 + - 25 + - 27 + - 29 + - 31 + - 33 + - 35 + - 37 + - 39 + - 41 + - 43 + - 45 + - 48 + - 52 + - 57 + - 64 + dim_head: 16 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: True + skip_connection: False + + +training: + batch_size: 4 + gradient_accumulation_steps: 1 + grad_clip: 1000.0 + instruments: ['noreverb', 'reverb'] + lr: 5.0e-5 + patience: 5 + reduce_factor: 0.75 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + + other_fix: False # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + + + +inference: + batch_size: 1 + dim_t: 871 + overlap_size: 19200 diff --git a/reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.ckpt b/reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..bc9d947ff83f741caa2504b673eed1983e6ac008 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26dda242bce4405555f2d6086d079fe8cc23f1f04e02e501d2689bfe3ece0489 +size 455864012 diff --git a/reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.yaml b/reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3a52977ab8c378560543e0dc4b6830eee04c21a4 --- /dev/null +++ b/reverb_echo_control/dereverb/dereverb_super_big_mbr_ep_346.yaml @@ -0,0 +1,76 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 8 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - dry + - No dry + lr: 4.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: dry + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 4 + dim_t: 801 + overlap_size: 17640 diff --git a/reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.ckpt b/reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5ebac2a6fd0b1e879b74243bf7bf6d2c17c4a305 --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee204fc59fa4111674536d47bd1ef3759acb9f7cf5a759ec4b867a828bb76c64 +size 170770820 diff --git a/reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.yaml b/reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c99b1ead72ed2ea1335e437c87a73ccf9a9e0e2c --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_bs_roformer_8_256dim_8depth.yaml @@ -0,0 +1,138 @@ +audio: + chunk_size: 352768 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 256 + depth: 8 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - No Reverb + - Reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: No Reverb + num_epochs: 1000 + num_steps: 7600 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17638 diff --git a/reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.ckpt b/reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..17a6af84d21d7c3cecf0c627811f42fee20b6d8e --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c38653aaa5e49f2f7b84dd3be2b6b679e0cbea23978e6b48389ee6f0a914768 +size 361499604 diff --git a/reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.yaml b/reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.yaml new file mode 100644 index 0000000000000000000000000000000000000000..305401363414e1be44ecd109979304663629c3de --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_bs_roformer_8_384dim_10depth.yaml @@ -0,0 +1,137 @@ +audio: + chunk_size: 352768 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 10 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 4 + dim_t: 801 + overlap_size: 17638 diff --git a/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.ckpt b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6ab261c7110f67d37ba5a6108012af71244cbf73 --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c3469d6b383b7ae0fff8aa6cb292eedf1eec132d3a960cadeecdec145123566d +size 438988833 diff --git a/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.yaml b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0764de2b0f67a6e9cc88c43b00eb2afcef39b30c --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_256dim_6depth.yaml @@ -0,0 +1,77 @@ +audio: + chunk_size: 352768 + dim_f: 1024 + dim_t: 256 + hop_length: 512 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 256 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 40000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 17638 diff --git a/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.ckpt b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f71612861aad2a454e2ca5268280d8bfafdc0af1 --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bd652149c50c6b816cdde6dbbb6ab2702e742fe25ec66592c6eb1ba9eb1a1ce +size 1708522165 diff --git a/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.yaml b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.yaml new file mode 100644 index 0000000000000000000000000000000000000000..03698129bdb0fc88407ef4737e0b51d8bbca2f71 --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_mel_band_roformer_8_512dim_12depth.yaml @@ -0,0 +1,77 @@ +audio: + chunk_size: 352768 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: noreverb + num_epochs: 1000 + num_steps: 1900 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17638 diff --git a/reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt b/reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..83776222e07a6b25db854916b814572b41764845 --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:29c311f2f3561c73f2084d59aa5dc2a077d907293bbbb1ce6548247ddb0671ff +size 489578757 diff --git a/reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.yaml b/reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ba2a25287daba8cb80169d0ecd1b87082ccc35ef --- /dev/null +++ b/reverb_echo_control/dereverb/deverb_mel_band_roformer_ep_27_sdr_10.4567.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 131584 + dim_f: 1024 + dim_t: 256 + hop_length: 512 + min_mean_abs: 0.001 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 + mixup: true + mixup_loudness_max: 1.5 + mixup_loudness_min: 0.5 + mixup_probs: !!python/tuple + - 0.2 + - 0.02 +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 6579 +model: + attn_dropout: 0.1 + depth: 12 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.1 + flash_attn: true + freq_transformer_depth: 1 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_bands: 60 + num_stems: 1 + sample_rate: 44100 + stereo: true + stft_hop_length: 512 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + batch_size: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - noreverb + - reverb + lr: 5.0e-05 + num_epochs: 1000 + num_steps: 1377 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: noreverb + use_amp: true diff --git a/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b2ec32944c1f4f9c802a1ad02d38682929c77a58 --- /dev/null +++ b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd2b737a394cfb80cd48cc9fcbaf89f5f4062f6b93066c2911617a06d8b7860a +size 835997896 diff --git a/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.yaml b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e12d8e53249bd5629448ae8a26afb32fa60b6c22 --- /dev/null +++ b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_10.0169.yaml @@ -0,0 +1,76 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 8 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - dry + - No dry + lr: 4.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 4 + dim_t: 801 + overlap_size: 17640 diff --git a/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5baa2b3d302d0ab984148ff1f42793220b29c7ca --- /dev/null +++ b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:396432f5af25992fe82d0286634bd879027c073721db6ab10199e75459708b9f +size 455862568 diff --git a/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.yaml b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..34e489695714f2e8268cb570cbc1b43b945de1d9 --- /dev/null +++ b/reverb_echo_control/dereverb_deecho/dereverb-echo_mel_band_roformer_sdr_13.4843_v2.yaml @@ -0,0 +1,64 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 8 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - dry + - No dry + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: dry + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false + use_amp: true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt b/reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2d495600912d8c596c4f37c1a902c9b918e934fd --- /dev/null +++ b/reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1596b1063238f487d54a0510a8c92cb28c000c803a271dd618ac49efc99ef3f7 +size 455776577 diff --git a/reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.yaml b/reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12a11d6b15a59acc5e99f68e39ca5f40ff4731f2 --- /dev/null +++ b/reverb_echo_control/dereverb_deecho/dereverb_echo_mbr_fused_0.5_v2_0.25_big_0.25_super.yaml @@ -0,0 +1,62 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 +model: + attn_dropout: 0.1 + depth: 8 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.1 + flash_attn: true + freq_transformer_depth: 1 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_bands: 60 + num_stems: 1 + sample_rate: 44100 + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + batch_size: 1 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 8 + instruments: + - dry + - other + lr: 4.0e-05 + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: dry + use_amp: true diff --git a/single_instrument/bass/HTDemucs4_FT_bass.th b/single_instrument/bass/HTDemucs4_FT_bass.th new file mode 100644 index 0000000000000000000000000000000000000000..b2516f0ba249a05c119eba7e3c52addc11785d41 --- /dev/null +++ b/single_instrument/bass/HTDemucs4_FT_bass.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e57c48e6b0e38af4f7118d7bd08c49f0a0c0edf7d09143bdd902ea0d237303e6 +size 84141271 diff --git a/single_instrument/bass/HTDemucs4_FT_bass.yaml b/single_instrument/bass/HTDemucs4_FT_bass.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cac91b35c6f74f25e84de93ae0fce0f49376b2d2 --- /dev/null +++ b/single_instrument/bass/HTDemucs4_FT_bass.yaml @@ -0,0 +1,102 @@ +audio: + chunk_size: 485100 + hop_length: 1024 + min_mean_abs: 0.001 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 +htdemucs: + bottom_channels: 512 + cac: true + channels: 48 + channels_time: null + context: 1 + context_enc: 0 + dconv_comp: 8 + dconv_depth: 2 + dconv_init: 1e-3 + dconv_mode: 3 + depth: 4 + emb_scale: 10 + emb_smooth: true + end_iters: 0 + freq_emb: 0.2 + growth: 2 + kernel_size: 8 + multi_freqs: [] + multi_freqs_depth: 3 + nfft: 4096 + norm_groups: 4 + norm_starts: 4 + num_subbands: 1 + rescale: 0.1 + rewrite: true + stride: 4 + t_auto_sparsity: false + t_cape_augment: true + t_cape_glob_loc_scale: + - 5000.0 + - 1.0 + - 1.4 + t_cape_mean_normalize: true + t_cross_first: false + t_dropout: 0.0 + t_emb: sin + t_gelu: true + t_global_window: 100 + t_group_norm: false + t_heads: 8 + t_hidden_scale: 4.0 + t_layer_scale: true + t_layers: 5 + t_lr: null + t_mask_random_seed: 42 + t_mask_type: diag + t_max_period: 10000.0 + t_max_positions: 10000 + t_norm_first: true + t_norm_in: true + t_norm_in_group: false + t_norm_out: true + t_sin_random_shift: 0 + t_sparse_attn_window: 400 + t_sparse_cross_attn: false + t_sparse_self_attn: false + t_sparsity: 0.95 + t_weight_decay: 0.0 + t_weight_pos_embed: 1.0 + time_stride: 2 + wiener_iters: 0 + wiener_residual: false +inference: + batch_size: 1 + overlap_size: 363825 +model: htdemucs +training: + batch_size: 8 + channels: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - drums + - bass + - other + - vocals + lr: 9.0e-05 + normalize: true + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + samplerate: 44100 + segment: 11 + shift: 1 + target_instrument: null + use_amp: true diff --git a/single_instrument/bass/model_jazz_scnet_bass_drums.ckpt b/single_instrument/bass/model_jazz_scnet_bass_drums.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2e5654b2add66e7f14e338f5e475e5b260c362c3 --- /dev/null +++ b/single_instrument/bass/model_jazz_scnet_bass_drums.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f89d89892782d39dabb356087cc36635dca2d52c13f27ba580ce0537c51030b +size 42458509 diff --git a/single_instrument/bass/model_jazz_scnet_bass_drums.yaml b/single_instrument/bass/model_jazz_scnet_bass_drums.yaml new file mode 100644 index 0000000000000000000000000000000000000000..883ef6c1f2c029c13a54d8cc47b38b059295f8d3 --- /dev/null +++ b/single_instrument/bass/model_jazz_scnet_bass_drums.yaml @@ -0,0 +1,83 @@ +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - drums + - bass + - piano + - other + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 10 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - piano + - other + lr: 5.0e-04 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 363825 + normalize: true diff --git a/single_instrument/drums_percussion/HTDemucs4_FT_drums.th b/single_instrument/drums_percussion/HTDemucs4_FT_drums.th new file mode 100644 index 0000000000000000000000000000000000000000..1d2f2cebac207dbb924b9011860f22621184d741 --- /dev/null +++ b/single_instrument/drums_percussion/HTDemucs4_FT_drums.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ba3fe64ae8ef66ac9a4857222ce48efbdc5eb3ad375cb79dd13debee5aaa4066 +size 84141271 diff --git a/single_instrument/drums_percussion/HTDemucs4_FT_drums.yaml b/single_instrument/drums_percussion/HTDemucs4_FT_drums.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cac91b35c6f74f25e84de93ae0fce0f49376b2d2 --- /dev/null +++ b/single_instrument/drums_percussion/HTDemucs4_FT_drums.yaml @@ -0,0 +1,102 @@ +audio: + chunk_size: 485100 + hop_length: 1024 + min_mean_abs: 0.001 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 +htdemucs: + bottom_channels: 512 + cac: true + channels: 48 + channels_time: null + context: 1 + context_enc: 0 + dconv_comp: 8 + dconv_depth: 2 + dconv_init: 1e-3 + dconv_mode: 3 + depth: 4 + emb_scale: 10 + emb_smooth: true + end_iters: 0 + freq_emb: 0.2 + growth: 2 + kernel_size: 8 + multi_freqs: [] + multi_freqs_depth: 3 + nfft: 4096 + norm_groups: 4 + norm_starts: 4 + num_subbands: 1 + rescale: 0.1 + rewrite: true + stride: 4 + t_auto_sparsity: false + t_cape_augment: true + t_cape_glob_loc_scale: + - 5000.0 + - 1.0 + - 1.4 + t_cape_mean_normalize: true + t_cross_first: false + t_dropout: 0.0 + t_emb: sin + t_gelu: true + t_global_window: 100 + t_group_norm: false + t_heads: 8 + t_hidden_scale: 4.0 + t_layer_scale: true + t_layers: 5 + t_lr: null + t_mask_random_seed: 42 + t_mask_type: diag + t_max_period: 10000.0 + t_max_positions: 10000 + t_norm_first: true + t_norm_in: true + t_norm_in_group: false + t_norm_out: true + t_sin_random_shift: 0 + t_sparse_attn_window: 400 + t_sparse_cross_attn: false + t_sparse_self_attn: false + t_sparsity: 0.95 + t_weight_decay: 0.0 + t_weight_pos_embed: 1.0 + time_stride: 2 + wiener_iters: 0 + wiener_residual: false +inference: + batch_size: 1 + overlap_size: 363825 +model: htdemucs +training: + batch_size: 8 + channels: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - drums + - bass + - other + - vocals + lr: 9.0e-05 + normalize: true + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + samplerate: 44100 + segment: 11 + shift: 1 + target_instrument: null + use_amp: true diff --git a/single_instrument/drums_percussion/model_jazz_scnet_piano_drums.ckpt b/single_instrument/drums_percussion/model_jazz_scnet_piano_drums.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d552c3d045a2c51f1546d286d39ceb10883c7446 --- /dev/null +++ b/single_instrument/drums_percussion/model_jazz_scnet_piano_drums.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d7fca79de2299382e72ad84d972755ed9830610b635ba55fba4f384f874a594 +size 42458509 diff --git a/single_instrument/drums_percussion/model_jazz_scnet_piano_drums.yaml b/single_instrument/drums_percussion/model_jazz_scnet_piano_drums.yaml new file mode 100644 index 0000000000000000000000000000000000000000..883ef6c1f2c029c13a54d8cc47b38b059295f8d3 --- /dev/null +++ b/single_instrument/drums_percussion/model_jazz_scnet_piano_drums.yaml @@ -0,0 +1,83 @@ +audio: + chunk_size: 485100 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - drums + - bass + - piano + - other + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 10 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - drums + - bass + - piano + - other + lr: 5.0e-04 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 363825 + normalize: true diff --git a/single_instrument/guitar/becruily_guitar.ckpt b/single_instrument/guitar/becruily_guitar.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9d3ea9afcc2d0451a33fa9fcb920ee53d17b0343 --- /dev/null +++ b/single_instrument/guitar/becruily_guitar.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83472bbf125774af5282d2e0b86df89eaf2dd45e8a4ec8d68e820ebf3e42a83c +size 45142183 diff --git a/single_instrument/guitar/becruily_guitar.yaml b/single_instrument/guitar/becruily_guitar.yaml new file mode 100644 index 0000000000000000000000000000000000000000..34a355b1b70b9e3a161458b6cf0a5a4adab4b3a9 --- /dev/null +++ b/single_instrument/guitar/becruily_guitar.yaml @@ -0,0 +1,65 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 4 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 1 + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 1.0 + instruments: + - Guitar + - Other + lr: 1.0 + patience: 1000 + reduce_factor: 0.95 + target_instrument: Guitar + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt b/single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..98d74da0b34a9fd746548ab608979f8bd076026e --- /dev/null +++ b/single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a81a87b46eed215c051579ac5cff3b139bfdfb9fa3bf4666df059ef87404d14 +size 276726667 diff --git a/single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.yaml b/single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6faecdc097c1ef2bc95ffdb982dfbdcf81c20d44 --- /dev/null +++ b/single_instrument/guitar/model_mel_band_roformer_ep_314_sdr_7.9825.yaml @@ -0,0 +1,74 @@ +audio: + chunk_size: 131584 + dim_f: 1024 + dim_t: 256 + hop_length: 512 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 192 + depth: 8 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 # Probably too big (requires a lot of memory for weights) + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 7 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - guitar + - others + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: guitar + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: true # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 6579 diff --git a/single_instrument/strings/gilliaan_bowedstrings_bs_v1.ckpt b/single_instrument/strings/gilliaan_bowedstrings_bs_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2af123913bf5657b9e21ca5ebd76eee9dfd00fc4 --- /dev/null +++ b/single_instrument/strings/gilliaan_bowedstrings_bs_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:282fabc28fb106edcc5e5e8383ea36559602fe53d0c406c736e08f58d66710fc +size 303490678 diff --git a/single_instrument/strings/gilliaan_bowedstrings_bs_v1.yaml b/single_instrument/strings/gilliaan_bowedstrings_bs_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8c8e340e0aa1078ac643225080fa7b9adf1578a8 --- /dev/null +++ b/single_instrument/strings/gilliaan_bowedstrings_bs_v1.yaml @@ -0,0 +1,128 @@ +audio: + chunk_size: 396900 # 44100 * 9 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False + skip_connection: False + +training: + batch_size: 3 + gradient_accumulation_steps: 4 + grad_clip: 1 + instruments: ['strings', 'other'] + lr: 5.0e-6 + patience: 2 + reduce_factor: 0.90 + target_instrument: null + num_epochs: 1000 + num_steps: 2100 # dataset / 3 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adamw + other_fix: False + use_amp: true + +inference: + batch_size: 2 + dim_t: 1251 + overlap_size: 19845 + normalize: false diff --git a/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.ckpt b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..65fa237173ab63980d5e176037e905439ed84b27 --- /dev/null +++ b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62579f748925893d0b1408e5a3319b1bbd8451a8f62d948e8438c72471f4574f +size 869726129 diff --git a/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.yaml b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..09a0398409f0ca796731acc4ae28958c98bc176d --- /dev/null +++ b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta1.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 317520 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + +training: + instruments: + - center + - wide + target_instrument: + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 256 + overlap_size: 15876 diff --git a/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.ckpt b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c3ef6c1f4bd66c469cb04ea3b1aabffbc1967d4c --- /dev/null +++ b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a77908f6e91d091e549525a4ad0053ce49a06ba4901a56659a76f864bc095502 +size 869726129 diff --git a/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.yaml b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..09a0398409f0ca796731acc4ae28958c98bc176d --- /dev/null +++ b/spatial/mono_stereo_duality/gilliaan_MonoStereo_Dual_Beta2.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 317520 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + +training: + instruments: + - center + - wide + target_instrument: + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 256 + overlap_size: 15876 diff --git a/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.ckpt b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..7fadec393b6e7b55177ad7e191df5ea5bdb45ba7 --- /dev/null +++ b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:334bc80ca5830e87e464694ff197c60372a2758c427d63588dc0cc2c1b3c885e +size 204495806 diff --git a/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.yaml b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..23c5efa25bc0fe76d1e75f0f4304efc7b50ca349 --- /dev/null +++ b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV1.yaml @@ -0,0 +1,128 @@ +audio: + chunk_size: 131072 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 1 + instruments: ['center', 'wide'] + lr: 1.0e-7 + patience: 3 + reduce_factor: 0.90 + target_instrument: center + num_epochs: 1000 + num_steps: 2000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adamw + other_fix: False # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1251 + overlap_size: 6554 + normalize: false diff --git a/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.ckpt b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c47fc8e8d42acd762fbbc07507da12520d62f4a9 --- /dev/null +++ b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72cbc991fc8fe96a9f6c33d5683649c662343a9e906e6c73189ff0f692ef9fe4 +size 204435682 diff --git a/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.yaml b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..23c5efa25bc0fe76d1e75f0f4304efc7b50ca349 --- /dev/null +++ b/spatial/phantom_center_wide/gilliaan_CenterExtractor_Single_StemV2.yaml @@ -0,0 +1,128 @@ +audio: + chunk_size: 131072 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 1 + instruments: ['center', 'wide'] + lr: 1.0e-7 + patience: 3 + reduce_factor: 0.90 + target_instrument: center + num_epochs: 1000 + num_steps: 2000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adamw + other_fix: False # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1251 + overlap_size: 6554 + normalize: false diff --git a/spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.ckpt b/spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d906c34a67462f59b6b48d06eed32b0447c7ce17 --- /dev/null +++ b/spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3817f96f5e446d07c91a731f37a00f5a3638a27d583b9390bb81c6cf001971c4 +size 214021674 diff --git a/spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.yaml b/spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c7688628e1307bbe67ff4671036c41e968c25dae --- /dev/null +++ b/spatial/phantom_center_wide/gilliaan_scnet_centerwide_dual.yaml @@ -0,0 +1,47 @@ +audio: + chunk_size: 154350 # 22050 * 7 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: ['center', 'wide'] + audio_channels: 2 + dims: [4, 64, 128, 256] + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: [0.230, 0.370, 0.400] + band_stride: [1, 4, 4] + band_kernel: [3, 4, 4] + conv_depths: [3, 3, 3] + compress: 4 + conv_kernel: 3 + num_dplayer: 8 + expand: 1 + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['center', 'wide'] + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1042 # dataset 2084 / 2 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adamw + lr: 1.0e-07 + normalize: false + other_fix: false + use_amp: true + +inference: + batch_size: 4 + dim_t: 256 + overlap_size: 115763 + normalize: false diff --git a/spatial/phantom_center_wide/model_mdx23c_ep_271_l1_freq_72.2383.ckpt b/spatial/phantom_center_wide/model_mdx23c_ep_271_l1_freq_72.2383.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d0268f2ad093d5b27ca484f20df986a025794a11 --- /dev/null +++ b/spatial/phantom_center_wide/model_mdx23c_ep_271_l1_freq_72.2383.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1227f8d97c7436004d03e46091427393abefebcc08ce53ef30082742c4e482f7 +size 437613512 diff --git a/spatial/phantom_center_wide/model_mdx23c_ep_271_l1_freq_72.2383.yaml b/spatial/phantom_center_wide/model_mdx23c_ep_271_l1_freq_72.2383.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b6d26909e869fd2fd2d2973591e38b6b9ef375a3 --- /dev/null +++ b/spatial/phantom_center_wide/model_mdx23c_ep_271_l1_freq_72.2383.yaml @@ -0,0 +1,48 @@ +audio: + chunk_size: 130560 + dim_f: 1024 + dim_t: 256 + hop_length: 512 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + act: gelu + bottleneck_factor: 4 + growth: 128 + norm: InstanceNorm + num_blocks_per_scale: 2 + num_channels: 128 + num_scales: 5 + num_subbands: 4 + scale: + - 2 + - 2 + +training: + batch_size: 2 + gradient_accumulation_steps: 3 + grad_clip: 0 + instruments: + - Similarity + - Difference + lr: 1.0 + patience: 15 + reduce_factor: 0.95 + target_instrument: Similarity + num_epochs: 1000 + num_steps: 2235 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 8 + dim_t: 256 + num_overlap: 8 + overlap_size: 114240 diff --git a/special/model_htdemucs_ep_21_sdr_13.6970.ckpt b/special/model_htdemucs_ep_21_sdr_13.6970.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..752ecd142025ffe7d1f8fc5a21f8b46dcded0005 --- /dev/null +++ b/special/model_htdemucs_ep_21_sdr_13.6970.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9970dca36a15c0d0cf3338d24592aa35469697dc3be4b07e9056f5d54b82185 +size 168122809 diff --git a/special/model_htdemucs_ep_21_sdr_13.6970.yaml b/special/model_htdemucs_ep_21_sdr_13.6970.yaml new file mode 100644 index 0000000000000000000000000000000000000000..55688b886ba98f12ceb3491cb19f79923212cf6e --- /dev/null +++ b/special/model_htdemucs_ep_21_sdr_13.6970.yaml @@ -0,0 +1,144 @@ +audio: + chunk_size: 132300 # samplerate * segment + min_mean_abs: 0.001 + hop_length: 1024 + +training: + batch_size: 5 + gradient_accumulation_steps: 3 + grad_clip: 0 + segment: 3 + shift: 1 + samplerate: 44100 + channels: 2 + normalize: true + instruments: ['similarity', 'difference'] + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + optimizer: prodigy + lr: 1.0 + patience: 80 + reduce_factor: 0.95 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: false # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: [0.2, 0.02] + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + overlap_size: 99225 + batch_size: 18 + +loss_multistft: + fft_sizes: + - 2048 + - 4096 + hop_sizes: + - 1024 + - 2048 + win_lengths: + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +model: htdemucs + +htdemucs: # see demucs/htdemucs.py for a detailed description + # Channels + channels: 48 + channels_time: + growth: 2 + # STFT + num_subbands: 1 + nfft: 4096 + wiener_iters: 0 + end_iters: 0 + wiener_residual: false + cac: true + # Main structure + depth: 4 + rewrite: true + # Frequency Branch + multi_freqs: [] + multi_freqs_depth: 3 + freq_emb: 0.2 + emb_scale: 10 + emb_smooth: true + # Convolutions + kernel_size: 8 + stride: 4 + time_stride: 2 + context: 1 + context_enc: 0 + # normalization + norm_starts: 4 + norm_groups: 4 + # DConv residual branch + dconv_mode: 3 + dconv_depth: 2 + dconv_comp: 8 + dconv_init: 1e-3 + # Before the Transformer + bottom_channels: 512 + # CrossTransformer + # ------ Common to all + # Regular parameters + t_layers: 5 + t_hidden_scale: 4.0 + t_heads: 8 + t_dropout: 0.0 + t_layer_scale: True + t_gelu: True + # ------------- Positional Embedding + t_emb: sin + t_max_positions: 10000 # for the scaled embedding + t_max_period: 10000.0 + t_weight_pos_embed: 1.0 + t_cape_mean_normalize: True + t_cape_augment: True + t_cape_glob_loc_scale: [5000.0, 1.0, 1.4] + t_sin_random_shift: 0 + # ------------- norm before a transformer encoder + t_norm_in: True + t_norm_in_group: False + # ------------- norm inside the encoder + t_group_norm: False + t_norm_first: True + t_norm_out: True + # ------------- optim + t_weight_decay: 0.0 + t_lr: + # ------------- sparsity + t_sparse_self_attn: False + t_sparse_cross_attn: False + t_mask_type: diag + t_mask_random_seed: 42 + t_sparse_attn_window: 400 + t_global_window: 100 + t_sparsity: 0.95 + t_auto_sparsity: False + # Cross Encoder First (False) + t_cross_first: False + # Weight init + rescale: 0.1 + diff --git a/special/model_scnet_ep_102_sdr_12.8941.ckpt b/special/model_scnet_ep_102_sdr_12.8941.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..dda6e7259532cfdf15fe3d0689fa9765a553cae9 --- /dev/null +++ b/special/model_scnet_ep_102_sdr_12.8941.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f372358803933b27f5c0d4273d333bee46c57741a043ca30f062f204afe5b24 +size 42422546 diff --git a/special/model_scnet_ep_102_sdr_12.8941.yaml b/special/model_scnet_ep_102_sdr_12.8941.yaml new file mode 100644 index 0000000000000000000000000000000000000000..36d273d48d5b4ba87d1029820f380c1bfd9eec8d --- /dev/null +++ b/special/model_scnet_ep_102_sdr_12.8941.yaml @@ -0,0 +1,102 @@ +audio: + chunk_size: 131072 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - difference + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 14 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - similarity + - difference + lr: 1.0 + patience: 300 + reduce_factor: 0.95 + target_instrument: difference + num_epochs: 1000 + num_steps: 500 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 16 + dim_t: 256 + overlap_size: 98304 + normalize: false diff --git a/special/model_scnet_ep_30_sdr_15.1291.ckpt b/special/model_scnet_ep_30_sdr_15.1291.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..05009401316ee8c3585b0ecb612ae825ea424b3f --- /dev/null +++ b/special/model_scnet_ep_30_sdr_15.1291.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f9a8671b338ef136afe0deb812737537264b1f42d8e5bc3fcc656e2c778a4e4 +size 42422170 diff --git a/special/model_scnet_ep_30_sdr_15.1291.yaml b/special/model_scnet_ep_30_sdr_15.1291.yaml new file mode 100644 index 0000000000000000000000000000000000000000..36d273d48d5b4ba87d1029820f380c1bfd9eec8d --- /dev/null +++ b/special/model_scnet_ep_30_sdr_15.1291.yaml @@ -0,0 +1,102 @@ +audio: + chunk_size: 131072 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - difference + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 14 + gradient_accumulation_steps: 4 + grad_clip: 0 + instruments: + - similarity + - difference + lr: 1.0 + patience: 300 + reduce_factor: 0.95 + target_instrument: difference + num_epochs: 1000 + num_steps: 500 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 16 + dim_t: 256 + overlap_size: 98304 + normalize: false diff --git a/vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt b/vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..0dd326fa1c5c65e52583803582f154a0359a55bd --- /dev/null +++ b/vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83bfe991cec4fbadde9f30d1f79cd5293ad0b1f936256be327bba5cbb4883374 +size 835982664 diff --git a/vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.yaml b/vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.yaml new file mode 100644 index 0000000000000000000000000000000000000000..657dbb166a9dcf3c9afb543aa462d694a015b7a2 --- /dev/null +++ b/vocal/aspiration_breath/aspiration_mel_band_roformer_less_aggr_sdr_18.1201.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 + mixup: false + mixup_loudness_max: 1.5 + mixup_loudness_min: 0.5 + mixup_probs: !!python/tuple + - 0.2 + - 0.02 +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 +model: + attn_dropout: 0.1 + depth: 8 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.1 + flash_attn: true + freq_transformer_depth: 1 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_bands: 60 + num_stems: 2 + sample_rate: 44100 + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + batch_size: 1 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 8 + instruments: + - aspiration + - other + lr: 4.0e-05 + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: null + use_amp: true diff --git a/vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.ckpt b/vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1282613860a16200f301a12c7b8ec67d050c63be --- /dev/null +++ b/vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e791258c866c6c8da66052693d8cc3b64f1f42c01e052dbdc570cd278380cc5 +size 835983746 diff --git a/vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.yaml b/vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.yaml new file mode 100644 index 0000000000000000000000000000000000000000..657dbb166a9dcf3c9afb543aa462d694a015b7a2 --- /dev/null +++ b/vocal/aspiration_breath/aspiration_mel_band_roformer_sdr_18.9845.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 + mixup: false + mixup_loudness_max: 1.5 + mixup_loudness_min: 0.5 + mixup_probs: !!python/tuple + - 0.2 + - 0.02 +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 +model: + attn_dropout: 0.1 + depth: 8 + dim: 256 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0.1 + flash_attn: true + freq_transformer_depth: 1 + heads: 8 + linear_transformer_depth: 0 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_bands: 60 + num_stems: 2 + sample_rate: 44100 + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + batch_size: 1 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 8 + instruments: + - aspiration + - other + lr: 4.0e-05 + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: null + use_amp: true diff --git a/vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.ckpt b/vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..281253d86a41b89a6a9687d35f8c18451f3814fe --- /dev/null +++ b/vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e104ce6f6733542f8356ac2e59f2b7ece49ebfd2713d1ff635806de5ceb483f +size 42457314 diff --git a/vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.yaml b/vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a6e45cafc4fbaceff80182c1d513942b37b9da03 --- /dev/null +++ b/vocal/choir_vocal/model_scnet_ep_36_sdr_5.4596.yaml @@ -0,0 +1,107 @@ +audio: + chunk_size: 131072 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - alto + - bass + - soprano + - tenor + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 9 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - alto + - bass + - soprano + - tenor + lr: 5.0e-4 + patience: 6 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 16 + dim_t: 256 + overlap_size: 0 + normalize: false diff --git a/vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.ckpt b/vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b37ed842b922550e23bc92f86117c449c94ec534 --- /dev/null +++ b/vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:14a668939243ab1551d2643820d4ff1a2b8f3c4ac1b9642718d5673405ea2f8b +size 42457314 diff --git a/vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.yaml b/vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a6e45cafc4fbaceff80182c1d513942b37b9da03 --- /dev/null +++ b/vocal/choir_vocal/model_scnet_ep_42_sdr_5.2559.yaml @@ -0,0 +1,107 @@ +audio: + chunk_size: 131072 # 44100 * 11 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + sources: + - alto + - bass + - soprano + - tenor + audio_channels: 2 + dims: + - 4 + - 32 + - 64 + - 128 + nfft: 4096 + hop_size: 1024 + win_size: 4096 + normalized: True + band_SR: + - 0.175 + - 0.392 + - 0.433 + band_stride: + - 1 + - 4 + - 16 + band_kernel: + - 3 + - 4 + - 16 + conv_depths: + - 3 + - 2 + - 1 + compress: 4 + conv_kernel: 3 + num_dplayer: 6 + expand: 1 + +training: + batch_size: 9 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - alto + - bass + - soprano + - tenor + lr: 5.0e-4 + patience: 6 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adamw8bit + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + +inference: + batch_size: 16 + dim_t: 256 + overlap_size: 0 + normalize: false diff --git a/vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt b/vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f4d51b38aad47f5ef1710bc4136a112e7ab594e0 --- /dev/null +++ b/vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ae8ae1a7d2fab0a7c405fee27ac8332702a31d689de58c9fb6f63b30410832b +size 630204333 diff --git a/vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.yaml b/vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7ed7b9fe1603eabb195d8d317a2244674b217638 --- /dev/null +++ b/vocal/duet_vocal/model_mel_band_roformer_ep_0_sdr_7.9319_fixed.yaml @@ -0,0 +1,110 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 2 + use_torch_checkpoint: True + skip_connection: False + +training: + batch_size: 1 + gradient_accumulation_steps: 10 + grad_clip: 1 + instruments: ["singer_1", "singer_2"] + patience: 4 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: prodigy + #optimizer: adamw + lr: 1.0 #e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +loss_multistft: + fft_sizes: + - 1024 + - 2048 + - 4096 + hop_sizes: + - 512 + - 1024 + - 2048 + win_lengths: + - 1024 + - 2048 + - 4096 + window: "hann_window" + scale: "mel" + n_bins: 128 + sample_rate: 44100 + perceptual_weighting: true + w_sc: 1.0 + w_log_mag: 1.0 + w_lin_mag: 0.0 + w_phs: 0.0 + mag_distance: "L1" + +#optimizer: + # betas: [0.8, 0.99] + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.2 + loudness_max: 1.5 + mixup: false # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: + !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.1 + mixup_loudness_max: 1.5 + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + +inference: + batch_size: 6 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.ckpt b/vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..be7ae9ab2d5ea1d9fb3550edb47fe29150ca9b11 --- /dev/null +++ b/vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4300ef0b230bf308fc8a606bbdcf38cd5002856b6dbe625b73576257f8641ae +size 102387933 diff --git a/vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.yaml b/vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.yaml new file mode 100644 index 0000000000000000000000000000000000000000..369871508b79b815c8fa15a887b1c1fa0ade81f3 --- /dev/null +++ b/vocal/speech_vocal/speech_bs_roformer_AliceN_sdr_20.4440.yaml @@ -0,0 +1,135 @@ +audio: + chunk_size: 882000 + dim_f: 1024 + dim_t: 801 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0059 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + mlp_expansion_factor: 4 + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + patience: 3 + reduce_factor: 0.8 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + scheduler: "ReduceLROnPlateau" + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: prodigy + lr: 1.0 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +optimizer: + weight_decay: 0.01 + safeguard_warmup: true + use_bias_correction: true + + +inference: + batch_size: 1 + dim_t: 2001 + overlap_size: 44100 + normalize: false diff --git a/vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.ckpt b/vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3cc0222e52a56683c84a8dbc826007b518050a2f --- /dev/null +++ b/vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2bd58827042b372f010d9e0a6bcae025e37b4f256f329066cf160c315558137 +size 435091173 diff --git a/vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.yaml b/vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9889301eceec52a85fb799161ab2aac5039734c5 --- /dev/null +++ b/vocal/speech_vocal/speech_mel_roformer_AliceN_sdr_16.5716.yaml @@ -0,0 +1,65 @@ +audio: + chunk_size: 882000 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 0.0003 + patience: 2 + reduce_factor: 0.95 + target_instrument: + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false + use_amp: true + + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 44100 diff --git a/vocal/vocal_extraction/BS-Roformer-Resurrection.ckpt b/vocal/vocal_extraction/BS-Roformer-Resurrection.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..14e223aece4474ca86627c065d73e725c4466902 --- /dev/null +++ b/vocal/vocal_extraction/BS-Roformer-Resurrection.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9dbfe5cb572e4ed32a15ec727d7bd06c8d7aba97509e6fda5bc008bb1e0b2dd5 +size 204510749 diff --git a/vocal/vocal_extraction/BS-Roformer-Resurrection.yaml b/vocal/vocal_extraction/BS-Roformer-Resurrection.yaml new file mode 100644 index 0000000000000000000000000000000000000000..52fb7559517581bf96f928c4f4e015ba733159c6 --- /dev/null +++ b/vocal/vocal_extraction/BS-Roformer-Resurrection.yaml @@ -0,0 +1,138 @@ +audio: + chunk_size: 785920 + dim_f: 1024 + dim_t: 1536 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['vocals', 'other'] + patience: 3 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1536 + overlap_size: 39296 + normalize: false diff --git a/vocal/vocal_extraction/HTDemucs4_FT_vocals_official.th b/vocal/vocal_extraction/HTDemucs4_FT_vocals_official.th new file mode 100644 index 0000000000000000000000000000000000000000..87e7befdc8b254d6cd7bfc8c2739e405783b7c3b --- /dev/null +++ b/vocal/vocal_extraction/HTDemucs4_FT_vocals_official.th @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3cf25b222c4eed7cd49dd8b2c9597d50c18bd154090f7b919cfa5f93cf22c49 +size 84141271 diff --git a/vocal/vocal_extraction/HTDemucs4_FT_vocals_official.yaml b/vocal/vocal_extraction/HTDemucs4_FT_vocals_official.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cac91b35c6f74f25e84de93ae0fce0f49376b2d2 --- /dev/null +++ b/vocal/vocal_extraction/HTDemucs4_FT_vocals_official.yaml @@ -0,0 +1,102 @@ +audio: + chunk_size: 485100 + hop_length: 1024 + min_mean_abs: 0.001 +augmentations: + enable: true + loudness: true + loudness_max: 1.5 + loudness_min: 0.5 +htdemucs: + bottom_channels: 512 + cac: true + channels: 48 + channels_time: null + context: 1 + context_enc: 0 + dconv_comp: 8 + dconv_depth: 2 + dconv_init: 1e-3 + dconv_mode: 3 + depth: 4 + emb_scale: 10 + emb_smooth: true + end_iters: 0 + freq_emb: 0.2 + growth: 2 + kernel_size: 8 + multi_freqs: [] + multi_freqs_depth: 3 + nfft: 4096 + norm_groups: 4 + norm_starts: 4 + num_subbands: 1 + rescale: 0.1 + rewrite: true + stride: 4 + t_auto_sparsity: false + t_cape_augment: true + t_cape_glob_loc_scale: + - 5000.0 + - 1.0 + - 1.4 + t_cape_mean_normalize: true + t_cross_first: false + t_dropout: 0.0 + t_emb: sin + t_gelu: true + t_global_window: 100 + t_group_norm: false + t_heads: 8 + t_hidden_scale: 4.0 + t_layer_scale: true + t_layers: 5 + t_lr: null + t_mask_random_seed: 42 + t_mask_type: diag + t_max_period: 10000.0 + t_max_positions: 10000 + t_norm_first: true + t_norm_in: true + t_norm_in_group: false + t_norm_out: true + t_sin_random_shift: 0 + t_sparse_attn_window: 400 + t_sparse_cross_attn: false + t_sparse_self_attn: false + t_sparsity: 0.95 + t_weight_decay: 0.0 + t_weight_pos_embed: 1.0 + time_stride: 2 + wiener_iters: 0 + wiener_residual: false +inference: + batch_size: 1 + overlap_size: 363825 +model: htdemucs +training: + batch_size: 8 + channels: 2 + coarse_loss_clip: true + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - drums + - bass + - other + - vocals + lr: 9.0e-05 + normalize: true + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: false + patience: 2 + q: 0.95 + reduce_factor: 0.95 + samplerate: 44100 + segment: 11 + shift: 1 + target_instrument: null + use_amp: true diff --git a/vocal/vocal_extraction/Kim_MelBandRoformer.ckpt b/vocal/vocal_extraction/Kim_MelBandRoformer.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e9269937826d8cedf1855096bc9c1d49298bb4f8 --- /dev/null +++ b/vocal/vocal_extraction/Kim_MelBandRoformer.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87201f4d31afb5bc79993230fc49446918425574db48c01c405e44f365c7559e +size 913106900 diff --git a/vocal/vocal_extraction/Kim_MelBandRoformer.yaml b/vocal/vocal_extraction/Kim_MelBandRoformer.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f19df3dc4667e7e725e8f0e9508c79c2d79e2be5 --- /dev/null +++ b/vocal/vocal_extraction/Kim_MelBandRoformer.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + min_mean_abs: 0.0 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 17640 +model: + attn_dropout: 0 + depth: 6 + dim: 384 + dim_freqs_in: 1025 + dim_head: 64 + ff_dropout: 0 + flash_attn: true + freq_transformer_depth: 1 + heads: 8 + mask_estimator_depth: 2 + multi_stft_hop_size: 147 + multi_stft_normalized: false + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + num_bands: 60 + num_stems: 1 + sample_rate: 44100 + stereo: true + stft_hop_length: 441 + stft_n_fft: 2048 + stft_normalized: false + stft_win_length: 2048 + time_transformer_depth: 1 +training: + augmentation: false + augmentation_loudness: false + augmentation_loudness_max: 0 + augmentation_loudness_min: 0 + augmentation_loudness_type: 1 + augmentation_mix: false + augmentation_type: null + batch_size: 4 + coarse_loss_clip: false + ema_momentum: 0.999 + grad_clip: 0 + gradient_accumulation_steps: 1 + instruments: + - vocals + - other + lr: 1.0e-05 + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + other_fix: true + patience: 2 + q: 0.95 + reduce_factor: 0.95 + target_instrument: vocals + use_amp: true + use_mp3_compress: false diff --git a/vocal/vocal_extraction/big_beta5e.ckpt b/vocal/vocal_extraction/big_beta5e.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a59473cd797f3c5fc6d6eb420d324145dc7c1104 --- /dev/null +++ b/vocal/vocal_extraction/big_beta5e.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32b876e1163716a9a007438b5a5107069586aa9b9ca653a5f63013b1edf6920c +size 1479749810 diff --git a/vocal/vocal_extraction/big_beta5e.yaml b/vocal/vocal_extraction/big_beta5e.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5df19f72cd2b2e9090b6dba59201835aa4e6d717 --- /dev/null +++ b/vocal/vocal_extraction/big_beta5e.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - vocals + - other + target_instrument: vocals + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/big_beta6.ckpt b/vocal/vocal_extraction/big_beta6.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..3e62cb6881f880ecadb2131661a7bffc1c651889 --- /dev/null +++ b/vocal/vocal_extraction/big_beta6.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f51cbb94b4ed5c36cb36fd2024236a8af3ed6886567981702ad6f094b2c6c820 +size 1557078584 diff --git a/vocal/vocal_extraction/big_beta6.yaml b/vocal/vocal_extraction/big_beta6.yaml new file mode 100644 index 0000000000000000000000000000000000000000..89456a257c1d0f22c0b157ca0e183b6733b3f452 --- /dev/null +++ b/vocal/vocal_extraction/big_beta6.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 529200 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 512 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1201 + overlap_size: 26460 diff --git a/vocal/vocal_extraction/big_beta6x.ckpt b/vocal/vocal_extraction/big_beta6x.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..8a45e7206470ce16a36e6c5c60726331d16221db --- /dev/null +++ b/vocal/vocal_extraction/big_beta6x.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e16d702f4e20f13d60b293541c1dea75cb4414a5846b36780e28ef70352a4e5c +size 1708527586 diff --git a/vocal/vocal_extraction/big_beta6x.yaml b/vocal/vocal_extraction/big_beta6x.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3c30f069c128ac5e7ab8482daaae858652f9f439 --- /dev/null +++ b/vocal/vocal_extraction/big_beta6x.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 529200 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1201 + overlap_size: 26460 diff --git a/vocal/vocal_extraction/big_beta7.ckpt b/vocal/vocal_extraction/big_beta7.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..dfb1e871f2703c66518d10374676aead2e4c2169 --- /dev/null +++ b/vocal/vocal_extraction/big_beta7.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9d68b9a8689a3500c45d3418a7811934e557760db07285f483088a0965b0eb88 +size 944675923 diff --git a/vocal/vocal_extraction/big_beta7.yaml b/vocal/vocal_extraction/big_beta7.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ccd25999f23841e17ed2a98100d3a1ac68615733 --- /dev/null +++ b/vocal/vocal_extraction/big_beta7.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 676935 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1201 + overlap_size: 33847 diff --git a/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt b/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2a346fe1f813eaf574a5e426eb004beb227b8026 --- /dev/null +++ b/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:54cf516f621f2f460bf660ed137e244b8931bf7a2ce85ddceecff816dbc4d668 +size 288724430 diff --git a/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.yaml b/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b492d88b145b26f15bc657d7ab66015ce465b5a5 --- /dev/null +++ b/vocal/vocal_extraction/bs_roformer_voc_hyperacev2.yaml @@ -0,0 +1,129 @@ +audio: + chunk_size: 960000 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0001 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: True + skip_connection: False + + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['vocals', 'instrument'] + lr: 1.0e-5 + patience: 5 + reduce_factor: 0.9 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + + +inference: + batch_size: 2 + dim_t: 1876 + overlap_size: 48000 diff --git a/vocal/vocal_extraction/bs_roformer_vocals_gabox.ckpt b/vocal/vocal_extraction/bs_roformer_vocals_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..cb4a0c8ef7ec4378b27e79a01eb491a2d699a535 --- /dev/null +++ b/vocal/vocal_extraction/bs_roformer_vocals_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18d58efe5e949e70fab11b875329af6d06ef11ccc29574bfe943fb57cc827f38 +size 639254584 diff --git a/vocal/vocal_extraction/bs_roformer_vocals_gabox.yaml b/vocal/vocal_extraction/bs_roformer_vocals_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c37ab724a16c904935f347a21e1ebb02ff748b78 --- /dev/null +++ b/vocal/vocal_extraction/bs_roformer_vocals_gabox.yaml @@ -0,0 +1,133 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 16 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/inst_fv7b.ckpt b/vocal/vocal_extraction/inst_fv7b.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..06903adc89757f37301f8386e7c03c9cba770e3b --- /dev/null +++ b/vocal/vocal_extraction/inst_fv7b.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cf9538616c835575eb07bef1fe4550b2406c852919a5644d22e9ae84a2ec6ba +size 913030778 diff --git a/vocal/vocal_extraction/inst_fv7b.yaml b/vocal/vocal_extraction/inst_fv7b.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e30949e959a05281c1b87bf8314353bf16b98ec1 --- /dev/null +++ b/vocal/vocal_extraction/inst_fv7b.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/instv7beta2.ckpt b/vocal/vocal_extraction/instv7beta2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..6fc66e39e0c0e766ab1479a213299e22d8df808a --- /dev/null +++ b/vocal/vocal_extraction/instv7beta2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11dcd18b2bca5ff189924145a52c9e1a145632c7659aac59b239221dca4f1703 +size 913026650 diff --git a/vocal/vocal_extraction/instv7beta2.yaml b/vocal/vocal_extraction/instv7beta2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e30949e959a05281c1b87bf8314353bf16b98ec1 --- /dev/null +++ b/vocal/vocal_extraction/instv7beta2.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/instv7beta3.ckpt b/vocal/vocal_extraction/instv7beta3.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..13bdd1d84fe6424a145c6ab381b094216899992e --- /dev/null +++ b/vocal/vocal_extraction/instv7beta3.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc33cc26ab805b057e47a753241bc17102564af4bc1c0e76364f3bb6a31c09d2 +size 913026650 diff --git a/vocal/vocal_extraction/instv7beta3.yaml b/vocal/vocal_extraction/instv7beta3.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e30949e959a05281c1b87bf8314353bf16b98ec1 --- /dev/null +++ b/vocal/vocal_extraction/instv7beta3.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/instv7plus.ckpt b/vocal/vocal_extraction/instv7plus.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..414302e59450aad73f3128193f0c490d8ea56e1f --- /dev/null +++ b/vocal/vocal_extraction/instv7plus.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7baf1aab873e534a186e2dcf6e64bebad2dbf3e2c49337729487294539118232 +size 913031195 diff --git a/vocal/vocal_extraction/instv7plus.yaml b/vocal/vocal_extraction/instv7plus.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e30949e959a05281c1b87bf8314353bf16b98ec1 --- /dev/null +++ b/vocal/vocal_extraction/instv7plus.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/kimmel_unwa_ft.ckpt b/vocal/vocal_extraction/kimmel_unwa_ft.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2912039564c3940b1fb51563a795e3f241841226 --- /dev/null +++ b/vocal/vocal_extraction/kimmel_unwa_ft.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6bd8d333880191254a6ef6be3cb0ffa4dda9d3282e36b0cce2e88a660e00d39 +size 913100690 diff --git a/vocal/vocal_extraction/kimmel_unwa_ft.yaml b/vocal/vocal_extraction/kimmel_unwa_ft.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e02996a0b3711eb4e8174cfa9cad0d2faec6bae4 --- /dev/null +++ b/vocal/vocal_extraction/kimmel_unwa_ft.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/kimmel_unwa_ft2.ckpt b/vocal/vocal_extraction/kimmel_unwa_ft2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..64a9c082838de39b001e92ac622efe3de0c810da --- /dev/null +++ b/vocal/vocal_extraction/kimmel_unwa_ft2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ed7b9e4c2eebbec7a7e5e8113058f7b68ba5e6048db8eaccfbbeb884c7884c0 +size 913100690 diff --git a/vocal/vocal_extraction/kimmel_unwa_ft2.yaml b/vocal/vocal_extraction/kimmel_unwa_ft2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e02996a0b3711eb4e8174cfa9cad0d2faec6bae4 --- /dev/null +++ b/vocal/vocal_extraction/kimmel_unwa_ft2.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/kimmel_unwa_ft3_prev.ckpt b/vocal/vocal_extraction/kimmel_unwa_ft3_prev.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..816aa053cbceb5323a1b6eb5b5c0717b9e0b7a66 --- /dev/null +++ b/vocal/vocal_extraction/kimmel_unwa_ft3_prev.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b10db0095c42576a32aeb3a1c8054af9dea0333e1c38477091d78316a007e52 +size 913090472 diff --git a/vocal/vocal_extraction/kimmel_unwa_ft3_prev.yaml b/vocal/vocal_extraction/kimmel_unwa_ft3_prev.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e02996a0b3711eb4e8174cfa9cad0d2faec6bae4 --- /dev/null +++ b/vocal/vocal_extraction/kimmel_unwa_ft3_prev.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/last_bs_roformer.ckpt b/vocal/vocal_extraction/last_bs_roformer.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e6f1d97debd68c9f88cf6c4e86678370f27bd2b6 --- /dev/null +++ b/vocal/vocal_extraction/last_bs_roformer.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:028ea94e354218fb4218b76233834a82e281661bcb6e613f6d04a7173b6455eb +size 204483448 diff --git a/vocal/vocal_extraction/last_bs_roformer.yaml b/vocal/vocal_extraction/last_bs_roformer.yaml new file mode 100644 index 0000000000000000000000000000000000000000..737bea1e47d00c234de25abf1947f0bc30258791 --- /dev/null +++ b/vocal/vocal_extraction/last_bs_roformer.yaml @@ -0,0 +1,129 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: True + skip_connection: False + + +training: + batch_size: 1 + gradient_accumulation_steps: 999 + grad_clip: 1 + instruments: ['vocals', 'other'] + lr: 1.0e-5 + patience: 1000000 + reduce_factor: 0.75 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: Fira + other_fix: True # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + use_torch_checkpoint: True + +inference: + batch_size: 6 + dim_t: 1251 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..380b1baf015018a1c4ae18ef1ade63a6ab76e441 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a64a27a672b457de23d9decd1fc7b58b0664a9f4f24bb43af154708e2ef07d2f +size 913090472 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.yaml b/vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8cb9e2b3d68e8e37b36e72f6f94d387f92379653 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocal_fullness_aname.yaml @@ -0,0 +1,54 @@ +audio: + chunk_size: 661500 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + target_instrument: vocals + use_amp: true + +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 33075 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_becruily.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_becruily.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..fda579c578c11b43fc8267f25fdbe0552ec36c1b --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_becruily.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a05961310cc55fbb901290c2e8be02682942f73522b6ac76bf2ec11e347ed95a +size 913107578 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_becruily.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_becruily.yaml new file mode 100644 index 0000000000000000000000000000000000000000..52215a606fcbf670c4f6584db7846d6af9dfd4b5 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_becruily.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 0.0005 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5b6fdc61e56b41297707de241f7c353a916c0494 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4dff354d81152d1b4321f6491f242c060919148239fbfe22a1015513de4a7fe +size 913026650 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv1_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..dccb5cbb2e7b8aa143e8cd8b803144646998abd5 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2888813aa5b519941fa8548efc5a4331d63c61909007eb17fe95c367be230196 +size 913026650 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv2_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f5b0031173c6b54221895b9d66e9553037777cd5 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d81446b34a7848446efde7898b25bdc32fe872c2393617acb5356649f7ea93 +size 913026650 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv3_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..16f6cdd40cf3df7b1fbbf3f3702c1f326cdd8f94 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a9657de5fd3ed87ad4fd1a9d2069743ecb33424836973ad0f3288e2a64e90bc +size 913026650 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv4_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..144ccb6ac58988f75e37cb1e16dd29d7ae754760 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ede0504ddc55cb44b966a8212dac75a364f8157974cc40c8e92b9f5d4f17ce2 +size 913026650 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv5_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..09ba0100d657171cb6f17292df03f215102f5bef --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25033d944288cb032fc51faab044bbd7f90bb81e82cada637ecdec699c2ff773 +size 913031195 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv6_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..0e898ed6634630dca5e5ad8156e1e4a2c6cbd0f5 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bb98488b80e7797e6f447c0dd8f89f2b7040eebc5e839b5a59871b7a3a19265 +size 913091027 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_fv7b_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..709869d299af3267439c266eede088a1c19b0745 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff802a67501fac70587c3ff4e8dbc89c2558e7d8911c92222dfea2aaac208517 +size 913026650 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.ckpt b/vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e3589e9ee8a2eba303868270993bbe81cc684625 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fac81dbebc0992503df55110d64d86c4fb74a1529527a819a253f3d20ef72bc1 +size 913031195 diff --git a/vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.yaml b/vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4afba4641e31905ec83d9852a0e9bfc74c5b7595 --- /dev/null +++ b/vocal/vocal_extraction/mel_band_roformer_vocals_v2_gabox.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: Vocals + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 + chunk_size: 352800 \ No newline at end of file diff --git a/vocal/vocal_extraction/melband_roformer_big_beta1.ckpt b/vocal/vocal_extraction/melband_roformer_big_beta1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..de4690f4f652d9ee11726628ea019013d8dab8e5 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59dfc7ecd486cb6276b00ab5618208fddda7d6bf5c1f0f6e3e769be9fa2801de +size 1479745022 diff --git a/vocal/vocal_extraction/melband_roformer_big_beta1.yaml b/vocal/vocal_extraction/melband_roformer_big_beta1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5df19f72cd2b2e9090b6dba59201835aa4e6d717 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta1.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - vocals + - other + target_instrument: vocals + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/melband_roformer_big_beta2.ckpt b/vocal/vocal_extraction/melband_roformer_big_beta2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4ac163ea7defcb5d325585026836b76a7cf8c340 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:756d87988f98c6333395d3621b18855d82a0faefea1dcd0c168108c1964dcde3 +size 1479745022 diff --git a/vocal/vocal_extraction/melband_roformer_big_beta2.yaml b/vocal/vocal_extraction/melband_roformer_big_beta2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5df19f72cd2b2e9090b6dba59201835aa4e6d717 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta2.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - vocals + - other + target_instrument: vocals + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/melband_roformer_big_beta3.ckpt b/vocal/vocal_extraction/melband_roformer_big_beta3.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c331fd1994de2eead2b76a82f7f191e20a7e2e83 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta3.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e67ca768e2fe3ef61a83702dcafa861df8a70b17a30262c4c00195e632afcd6b +size 1479738496 diff --git a/vocal/vocal_extraction/melband_roformer_big_beta3.yaml b/vocal/vocal_extraction/melband_roformer_big_beta3.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5df19f72cd2b2e9090b6dba59201835aa4e6d717 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta3.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - vocals + - other + target_instrument: vocals + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/melband_roformer_big_beta4.ckpt b/vocal/vocal_extraction/melband_roformer_big_beta4.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..62cb0ed8e023394924b1108a5335daa8b91d78d4 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta4.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:700a9bd3831d4f7f44cc0019b238774e31045bcbc361fbb69235535c40fc1454 +size 1574477088 diff --git a/vocal/vocal_extraction/melband_roformer_big_beta4.yaml b/vocal/vocal_extraction/melband_roformer_big_beta4.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9b4ee013ca1a7a2e2eb82099699f6abef44258c0 --- /dev/null +++ b/vocal/vocal_extraction/melband_roformer_big_beta4.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - vocals + - other + target_instrument: vocals + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.ckpt b/vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1d4b892da79c875b3b3028f9f4d2504ebafe72e1 --- /dev/null +++ b/vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b84f37e8d444c8cb30c79d77f613a41c05868ff9c9ac6c7049c00aefae115aa +size 639331213 diff --git a/vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.yaml b/vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c37ab724a16c904935f347a21e1ebb02ff748b78 --- /dev/null +++ b/vocal/vocal_extraction/model_bs_roformer_ep_317_sdr_12.9755.yaml @@ -0,0 +1,133 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 16 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.ckpt b/vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..929fbf0bc57f788a3d6b758ee4feedd61976298b --- /dev/null +++ b/vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6c94864adfb73bbb0ca58ec14d58dd0b364549e9fb61433ae51916f3e2f8d0b +size 639317465 diff --git a/vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.yaml b/vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c37ab724a16c904935f347a21e1ebb02ff748b78 --- /dev/null +++ b/vocal/vocal_extraction/model_bs_roformer_ep_368_sdr_12.9628.yaml @@ -0,0 +1,133 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 16 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/model_bsrofomer_essid_vocals.ckpt b/vocal/vocal_extraction/model_bsrofomer_essid_vocals.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4537f7c059ee1aedf28f165dca8523affe73f834 --- /dev/null +++ b/vocal/vocal_extraction/model_bsrofomer_essid_vocals.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f4af693f8fd546f218f2df63514e063479ff29e445f21e54f79c33685d64d47 +size 639330302 diff --git a/vocal/vocal_extraction/model_bsrofomer_essid_vocals.yaml b/vocal/vocal_extraction/model_bsrofomer_essid_vocals.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4b363eeae53fb71108f69c8c8cdedf2f3e595fec --- /dev/null +++ b/vocal/vocal_extraction/model_bsrofomer_essid_vocals.yaml @@ -0,0 +1,126 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 512 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 4 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/model_mdx23c_ep_0_sdr_14.6632.ckpt b/vocal/vocal_extraction/model_mdx23c_ep_0_sdr_14.6632.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..c9a94f6ee63078f4716217a12fa92c2e3bbe2672 --- /dev/null +++ b/vocal/vocal_extraction/model_mdx23c_ep_0_sdr_14.6632.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f75a7e9cf05664fd265e14c7225d85c15c6498b98c55a7ddbfd44fbd0ec13841 +size 448103606 diff --git a/vocal/vocal_extraction/model_mdx23c_ep_0_sdr_14.6632.yaml b/vocal/vocal_extraction/model_mdx23c_ep_0_sdr_14.6632.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8793f19ef81dc6de9fc02f5eb4dcec03833d0aa7 --- /dev/null +++ b/vocal/vocal_extraction/model_mdx23c_ep_0_sdr_14.6632.yaml @@ -0,0 +1,97 @@ +audio: + chunk_size: 261120 + dim_f: 4096 + dim_t: 256 + hop_length: 1024 + n_fft: 8192 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + act: gelu + bottleneck_factor: 4 + growth: 128 + norm: InstanceNorm + num_blocks_per_scale: 2 + num_channels: 128 + num_scales: 5 + num_subbands: 4 + scale: + - 2 + - 2 + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + read_metadata_procs: 8 # Number of processes to use during metadata reading for dataset. Can speed up metadata generation + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +augmentations: + enable: false # enable or disable all augmentations (to fast disable if needed) + loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max) + loudness_min: 0.5 + loudness_max: 1.5 + mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3) + mixup_probs: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02) + - 0.2 + - 0.02 + mixup_loudness_min: 0.5 + mixup_loudness_max: 1.5 + + # apply mp3 compression to mixture only (emulate downloading mp3 from internet) + mp3_compression_on_mixture: 0.01 + mp3_compression_on_mixture_bitrate_min: 32 + mp3_compression_on_mixture_bitrate_max: 320 + mp3_compression_on_mixture_backend: "lameenc" + + all: + channel_shuffle: 0.5 # Set 0 or lower to disable + random_inverse: 0.1 # inverse track (better lower probability) + random_polarity: 0.5 # polarity change (multiply waveform to -1) + mp3_compression: 0.01 + mp3_compression_min_bitrate: 32 + mp3_compression_max_bitrate: 320 + mp3_compression_backend: "lameenc" + + vocals: + pitch_shift: 0.1 + pitch_shift_min_semitones: -5 + pitch_shift_max_semitones: 5 + seven_band_parametric_eq: 0.25 + seven_band_parametric_eq_min_gain_db: -9 + seven_band_parametric_eq_max_gain_db: 9 + tanh_distortion: 0.1 + tanh_distortion_min: 0.1 + tanh_distortion_max: 0.7 + other: + pitch_shift: 0.1 + pitch_shift_min_semitones: -4 + pitch_shift_max_semitones: 4 + gaussian_noise: 0.1 + gaussian_noise_min_amplitude: 0.001 + gaussian_noise_max_amplitude: 0.015 + time_stretch: 0.01 + time_stretch_min_rate: 0.8 + time_stretch_max_rate: 1.25 + +inference: + batch_size: 1 + dim_t: 256 + num_overlap: 4 + overlap_size: 195840 \ No newline at end of file diff --git a/vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt b/vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..1cb36d54103c4fdd30f7e8a089164cba50cfb0b0 --- /dev/null +++ b/vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21b9d0958e35b8ebfbe2afe69bbd5444e5ffe2f5d80ae0d583b833d2f3c0d139 +size 1007816988 diff --git a/vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.yaml b/vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b25fd29f1ef3e5c40d88fb5c566dadea893870fd --- /dev/null +++ b/vocal/vocal_extraction/model_mel_band_roformer_ep_3005_sdr_11.4360.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 384 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 9 + gradient_accumulation_steps: 8 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 4.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/model_vocals.ckpt b/vocal/vocal_extraction/model_vocals.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..198da442eeb58a6cb5cb514ce350a9cbb771f5cb --- /dev/null +++ b/vocal/vocal_extraction/model_vocals.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e89b915fb0d4b1f23a9ebed8f499ada3c212256fa6f449e1b2c16412eeb3c38e +size 725806842 diff --git a/vocal/vocal_extraction/model_vocals.yaml b/vocal/vocal_extraction/model_vocals.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7f252626dabb4d636f8178fd907ca491cf48404a --- /dev/null +++ b/vocal/vocal_extraction/model_vocals.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 320 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0.0 + instruments: + - vocals + - other + lr: 1.0 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: true # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.ckpt b/vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..873781d9bbbfe3ae2430615d148a00dec77b5475 --- /dev/null +++ b/vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9ce706b49cebf0af018590d8deb47ad5434987bf8f7bd3a87a4e5e8c30acb26 +size 134917909 diff --git a/vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.yaml b/vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a8dd3e3556e1578de7f529cc45d98430e4ab5c6d --- /dev/null +++ b/vocal/vocal_extraction/model_vocals_mel_band_roformer_sdr_8.42.yaml @@ -0,0 +1,71 @@ +audio: + chunk_size: 131584 + dim_f: 1024 + dim_t: 256 + hop_length: 512 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 + +model: + dim: 192 + depth: 8 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0.1 + ff_dropout: 0.1 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 512 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 1 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 7 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 5.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 6579 diff --git a/vocal/vocal_extraction/voc_fv7.ckpt b/vocal/vocal_extraction/voc_fv7.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e33f4329116367aea94e31f0f4bca29098889e7f --- /dev/null +++ b/vocal/vocal_extraction/voc_fv7.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77b9e4b54802d670c02daceb30c61ec825dc54ca6c29c34b03cdd5e9f78382b6 +size 489571079 diff --git a/vocal/vocal_extraction/voc_fv7.yaml b/vocal/vocal_extraction/voc_fv7.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e7a5af029ba415abb5aec0a960a27915eb7e3b49 --- /dev/null +++ b/vocal/vocal_extraction/voc_fv7.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/vocfv7beta2.ckpt b/vocal/vocal_extraction/vocfv7beta2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..872c0e5ec84bdcefb1ed67affa0ad60113236ac2 --- /dev/null +++ b/vocal/vocal_extraction/vocfv7beta2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85bcaa03568f96e56b245302c177658e8cb83aa6e2f4ae260e67202db9d56123 +size 913030778 diff --git a/vocal/vocal_extraction/vocfv7beta2.yaml b/vocal/vocal_extraction/vocfv7beta2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e30949e959a05281c1b87bf8314353bf16b98ec1 --- /dev/null +++ b/vocal/vocal_extraction/vocfv7beta2.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_extraction/vocfv7beta3.ckpt b/vocal/vocal_extraction/vocfv7beta3.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..eea302148fec03883ca472dee62ca4006aaaf84e --- /dev/null +++ b/vocal/vocal_extraction/vocfv7beta3.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45463b0051651b7ed15677cc41f21bd384d66796938f46955617bf6f1d661988 +size 913030778 diff --git a/vocal/vocal_extraction/vocfv7beta3.yaml b/vocal/vocal_extraction/vocfv7beta3.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e30949e959a05281c1b87bf8314353bf16b98ec1 --- /dev/null +++ b/vocal/vocal_extraction/vocfv7beta3.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: Vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt b/vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..201059f3d93490e5cc91f20ea0bb74ae7c0dd20a --- /dev/null +++ b/vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cf11736d1b42a11ae55d8299316585921477dd2a671b24b663660846ca9861b +size 527119779 diff --git a/vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.yaml b/vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e056cf10a3a7dc497f0e770949081db0ba22d6f9 --- /dev/null +++ b/vocal/vocal_gender_chorus/bs_roformer_male_female_by_aufr33_sdr_7.2889.yaml @@ -0,0 +1,125 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - male + - female + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..992076a4a199b9e0725dc3f91b667ab6624fc80f --- /dev/null +++ b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:755e3641645687c618a448605ba71c30abd3be1b1a8861e5bfa12362a85c609a +size 527121477 diff --git a/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.yaml b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d154c3ee888a291651151ef96ad1be5ed13d2224 --- /dev/null +++ b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_146_sdr_23.8613.yaml @@ -0,0 +1,126 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - male + - female + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..4f9492413f4671a4c218f73cd6df2963e6b75fb8 --- /dev/null +++ b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:123c00786bdbc6bd462dddb35cd21fd6ae99ab8319f93f63a8abc1012e593d94 +size 527121477 diff --git a/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.yaml b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e056cf10a3a7dc497f0e770949081db0ba22d6f9 --- /dev/null +++ b/vocal/vocal_gender_chorus/model_chorus_bs_roformer_ep_267_sdr_24.1275.yaml @@ -0,0 +1,125 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 801 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0.0 + ff_dropout: 0.0 + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - male + - female + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 801 + overlap_size: 17640 diff --git a/vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.ckpt b/vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..23b7fc62506629fc0ac29c1551ee7045b805dfa5 --- /dev/null +++ b/vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16311025a5133ae6411760ccfe9e3e66b31a01d9d8bec0a03fa7ec4bedac7a15 +size 204483033 diff --git a/vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.yaml b/vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9823f92202c5917c4f5d961c72543c4fda940d0e --- /dev/null +++ b/vocal/vocal_instrumental_dual/BS-Roformer-Resurrection-Inst.yaml @@ -0,0 +1,138 @@ +audio: + chunk_size: 749259 + dim_f: 1024 + dim_t: 1700 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['vocals', 'other'] + patience: 3 + reduce_factor: 0.95 + target_instrument: other + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1700 + overlap_size: 37463 + normalize: false diff --git a/vocal/vocal_instrumental_dual/INSTV10.ckpt b/vocal/vocal_instrumental_dual/INSTV10.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5778b058fb0b07ce9c2ae2be090f2e8425bcfb9e --- /dev/null +++ b/vocal/vocal_instrumental_dual/INSTV10.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6eb987c3fd93c66adfea2e4284d1da668e80a98f75b6436b4be454780a728182 +size 913026650 diff --git a/vocal/vocal_instrumental_dual/INSTV10.yaml b/vocal/vocal_instrumental_dual/INSTV10.yaml new file mode 100644 index 0000000000000000000000000000000000000000..da6e2bc08e1301b754a246cf507c472468f9c164 --- /dev/null +++ b/vocal/vocal_instrumental_dual/INSTV10.yaml @@ -0,0 +1,73 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.00 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: true + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 250 + multi_stft_normalized: false + use_torch_checkpoint: true + +training: + batch_size: 1 + gradient_accumulation_steps: 999999999999999999999999 + grad_clip: 0 + instruments: + - other + - vocals + lr: 0.00001 + patience: 100000000 + reduce_factor: 0.95 + target_instrument: other + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 17640 diff --git a/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ.ckpt b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d19b34ba845c464fd607a4ce4c9dd3f794d0bd23 --- /dev/null +++ b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d51472769e34a2501cd1da782346a3212555c3a5619fc2c53507445528d816 +size 448101203 diff --git a/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ.yaml b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ.yaml new file mode 100644 index 0000000000000000000000000000000000000000..562448244ebf8692c728a478b0b6bc39b50ea2fd --- /dev/null +++ b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ.yaml @@ -0,0 +1,44 @@ +audio: + chunk_size: 261120 + dim_f: 4096 + dim_t: 256 + hop_length: 1024 + n_fft: 8192 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 +model: + act: gelu + bottleneck_factor: 4 + growth: 128 + norm: InstanceNorm + num_blocks_per_scale: 2 + num_channels: 128 + num_scales: 5 + num_subbands: 4 + scale: + - 2 + - 2 +training: + batch_size: 6 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + augmentation: 1 + augmentation_type: simple1 + augmentation_mix: true + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 +inference: + batch_size: 1 + dim_t: 256 + num_overlap: 8 + overlap_size: 228480 \ No newline at end of file diff --git a/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.ckpt b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..486c55bd1538c5471f02f9e08af5ceb8baec08e2 --- /dev/null +++ b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d960d8e40a458120412c1bd807e013d2dbca7b959cc9da2bbcb0eb203d1daea +size 448093770 diff --git a/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.yaml b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..562448244ebf8692c728a478b0b6bc39b50ea2fd --- /dev/null +++ b/vocal/vocal_instrumental_dual/MDX23C-8KFFT-InstVoc_HQ_2.yaml @@ -0,0 +1,44 @@ +audio: + chunk_size: 261120 + dim_f: 4096 + dim_t: 256 + hop_length: 1024 + n_fft: 8192 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 +model: + act: gelu + bottleneck_factor: 4 + growth: 128 + norm: InstanceNorm + num_blocks_per_scale: 2 + num_channels: 128 + num_scales: 5 + num_subbands: 4 + scale: + - 2 + - 2 +training: + batch_size: 6 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: null + num_epochs: 1000 + num_steps: 1000 + augmentation: 1 + augmentation_type: simple1 + augmentation_mix: true + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 +inference: + batch_size: 1 + dim_t: 256 + num_overlap: 8 + overlap_size: 228480 \ No newline at end of file diff --git a/vocal/vocal_instrumental_dual/MDX23C_D1581.ckpt b/vocal/vocal_instrumental_dual/MDX23C_D1581.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..62aeea1ec485760c60a560dbcc87e44995964d23 --- /dev/null +++ b/vocal/vocal_instrumental_dual/MDX23C_D1581.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7d4d2137f12764950205b095da20032fef1d41f077bacc8582f20ed40e8cb28 +size 183379219 diff --git a/vocal/vocal_instrumental_dual/MDX23C_D1581.yaml b/vocal/vocal_instrumental_dual/MDX23C_D1581.yaml new file mode 100644 index 0000000000000000000000000000000000000000..86c9bff9c6a364abb9369ff4c46cd0afdf3c89a0 --- /dev/null +++ b/vocal/vocal_instrumental_dual/MDX23C_D1581.yaml @@ -0,0 +1,37 @@ +audio: + chunk_size: 260096 + dim_f: 4096 + dim_t: 256 + hop_length: 2048 + n_fft: 12288 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.001 +model: + act: gelu + bottleneck_factor: 4 + growth: 64 + norm: InstanceNorm + num_blocks_per_scale: 2 + num_channels: 128 + num_scales: 5 + num_subbands: 4 + scale: + - 2 + - 2 + name: epoch_10.ckpt +training: + batch_size: 16 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 5.0e-05 + target_instrument: null + num_epochs: 100 + num_steps: 1000 +inference: + batch_size: 1 + dim_t: 256 + num_overlap: 8 + overlap_size: 227584 diff --git a/vocal/vocal_instrumental_dual/becruily_deux.ckpt b/vocal/vocal_instrumental_dual/becruily_deux.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..92b67dd153034ede8c0080310cb90ec9dd65af21 --- /dev/null +++ b/vocal/vocal_instrumental_dual/becruily_deux.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10255c02295bf3e3865d4ee50ff752d7b19b124ed5fd93b147babc4333eda3aa +size 435006815 diff --git a/vocal/vocal_instrumental_dual/becruily_deux.yaml b/vocal/vocal_instrumental_dual/becruily_deux.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c04a065bfbd1ae2d8e8ba5ba11231558c79f3eab --- /dev/null +++ b/vocal/vocal_instrumental_dual/becruily_deux.yaml @@ -0,0 +1,64 @@ +audio: + chunk_size: 573300 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - Vocals + - Instrumental + lr: 0.0001 + patience: 2 + reduce_factor: 0.95 + target_instrument: + num_epochs: 1000 + num_steps: 1000 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adamw + other_fix: false + use_amp: true + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 28665 diff --git a/vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.ckpt b/vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..285c18b708283179feaad81ea30f1bd6fb467c14 --- /dev/null +++ b/vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d28b3a422531e1ca482d29302a5b67614eb45c09d72ce954258011ce24ef0919 +size 204483033 diff --git a/vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.yaml b/vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3b474054b768075a1ac880991c7de998c3e52804 --- /dev/null +++ b/vocal/vocal_instrumental_dual/bs_roformer_instrumental_resurrection_gabox.yaml @@ -0,0 +1,135 @@ +audio: + chunk_size: 749259 + dim_f: 1024 + dim_t: 1700 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['vocals', 'other'] + patience: 3 + reduce_factor: 0.95 + target_instrument: other + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1700 + overlap_size: 37463 + normalize: false diff --git a/vocal/vocal_instrumental_dual/duality_v1.ckpt b/vocal/vocal_instrumental_dual/duality_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..a200aeb7a350ff42686f2933c050c2c5264a521d --- /dev/null +++ b/vocal/vocal_instrumental_dual/duality_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69c244e76d7142a948f0d41c67e293112ba84f12a28c89d24b22a3fecd9ef79e +size 913043907 diff --git a/vocal/vocal_instrumental_dual/duality_v1.yaml b/vocal/vocal_instrumental_dual/duality_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b2a83eb64139f5672c97900d94334c0fa8aafd46 --- /dev/null +++ b/vocal/vocal_instrumental_dual/duality_v1.yaml @@ -0,0 +1,69 @@ +audio: + chunk_size: 661500 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + batch_size: 4 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: vocals + num_epochs: 1000 + num_steps: 1000 + augmentation: false + augmentation_type: null + use_mp3_compress: false + augmentation_mix: false + augmentation_loudness: false + augmentation_loudness_type: 1 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true + use_amp: true +inference: + batch_size: 4 + dim_t: 1101 + overlap_size: 33075 diff --git a/vocal/vocal_instrumental_dual/essid_mel_vocal_v1.ckpt b/vocal/vocal_instrumental_dual/essid_mel_vocal_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..56507bfc9b81d032a51d6b72e5a11b3b40cbd407 --- /dev/null +++ b/vocal/vocal_instrumental_dual/essid_mel_vocal_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d4c8f69c52adce7b2fd3602fc004b27b4305700b70232b461fb3bb15400c3f2 +size 913106900 diff --git a/vocal/vocal_instrumental_dual/essid_mel_vocal_v1.yaml b/vocal/vocal_instrumental_dual/essid_mel_vocal_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12642216adf6fde4bf38ea0ee3ace87b3affee49 --- /dev/null +++ b/vocal/vocal_instrumental_dual/essid_mel_vocal_v1.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: other + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 17640 diff --git a/vocal/vocal_instrumental_dual/essid_mel_vocal_v2.ckpt b/vocal/vocal_instrumental_dual/essid_mel_vocal_v2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..5ab6a5b38a412ae9829395c4d37dd5039c1ab6e3 --- /dev/null +++ b/vocal/vocal_instrumental_dual/essid_mel_vocal_v2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33ef2fc4c87d4d1d02614027643246ff287f57e8fad497138c3b8a2574a2bed5 +size 913106900 diff --git a/vocal/vocal_instrumental_dual/essid_mel_vocal_v2.yaml b/vocal/vocal_instrumental_dual/essid_mel_vocal_v2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12642216adf6fde4bf38ea0ee3ace87b3affee49 --- /dev/null +++ b/vocal/vocal_instrumental_dual/essid_mel_vocal_v2.yaml @@ -0,0 +1,72 @@ +audio: + chunk_size: 352800 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + batch_size: 1 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: + - vocals + - other + lr: 1.0e-05 + patience: 2 + reduce_factor: 0.95 + target_instrument: other + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: null + use_mp3_compress: false # Deprecated + augmentation_mix: false # Mix several stems of the same type with some probability + augmentation_loudness: false # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0 + augmentation_loudness_max: 0 + q: 0.95 + coarse_loss_clip: false + ema_momentum: 0.999 + optimizer: adam + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 1 + dim_t: 256 + overlap_size: 17640 diff --git a/vocal/vocal_instrumental_dual/inst_v1_plus_test.ckpt b/vocal/vocal_instrumental_dual/inst_v1_plus_test.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..ba0efaf8a870d3f5921d78c2fe7107ebab72c7e6 --- /dev/null +++ b/vocal/vocal_instrumental_dual/inst_v1_plus_test.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25fe6ab4db95a6c20468f6c082f9bfc30904b8727ed8d069110bd0960da4879a +size 913090472 diff --git a/vocal/vocal_instrumental_dual/inst_v1_plus_test.yaml b/vocal/vocal_instrumental_dual/inst_v1_plus_test.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d272da0806584a19c2b51517819911361b3c8adf --- /dev/null +++ b/vocal/vocal_instrumental_dual/inst_v1_plus_test.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - other + - vocals + target_instrument: other + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_instrumental_dual/inst_v1e.ckpt b/vocal/vocal_instrumental_dual/inst_v1e.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..f8370873342b453fb65d96b2fe069cbc0f8130e9 --- /dev/null +++ b/vocal/vocal_instrumental_dual/inst_v1e.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df2bcdb8838b88264f5381dbb0ccd84a9926c9775cf548c34d8846f5cd20fe96 +size 913102724 diff --git a/vocal/vocal_instrumental_dual/inst_v1e.yaml b/vocal/vocal_instrumental_dual/inst_v1e.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d272da0806584a19c2b51517819911361b3c8adf --- /dev/null +++ b/vocal/vocal_instrumental_dual/inst_v1e.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - other + - vocals + target_instrument: other + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_instrumental_dual/inst_v1e_plus.ckpt b/vocal/vocal_instrumental_dual/inst_v1e_plus.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..d96e91089fa7a07f5c4052f55449fd6ed4047ce2 --- /dev/null +++ b/vocal/vocal_instrumental_dual/inst_v1e_plus.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a4ddba739f0352407fb6e18b29206b82318ec427fe37fcedb0f83241e4e15fb +size 913090472 diff --git a/vocal/vocal_instrumental_dual/inst_v1e_plus.yaml b/vocal/vocal_instrumental_dual/inst_v1e_plus.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d272da0806584a19c2b51517819911361b3c8adf --- /dev/null +++ b/vocal/vocal_instrumental_dual/inst_v1e_plus.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - other + - vocals + target_instrument: other + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_inst_v1.ckpt b/vocal/vocal_instrumental_dual/melband_roformer_inst_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..0a5775dcf471eae22321cf27576da3155bd773e9 --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_inst_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f88d96958b2b7dec32286b0ced00bbcbd37e28741cad9038758b1eaf9b5c057 +size 913100690 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_inst_v1.yaml b/vocal/vocal_instrumental_dual/melband_roformer_inst_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d272da0806584a19c2b51517819911361b3c8adf --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_inst_v1.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - other + - vocals + target_instrument: other + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_inst_v2.ckpt b/vocal/vocal_instrumental_dual/melband_roformer_inst_v2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e59c3365c4976b3c225f0cf9ab34342531dc6955 --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_inst_v2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd19766620f7d6f58fdf7aaada7e89907fe41bc64490ce3faa9a6dab15d6e1f2 +size 1574477088 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_inst_v2.yaml b/vocal/vocal_instrumental_dual/melband_roformer_inst_v2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0a0c864d9190446819059730665d2d258d44bf1f --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_inst_v2.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 1101 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 3 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - other + - vocals + target_instrument: other + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.ckpt b/vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..2b99b0c8e53949042dce6f1db2300642212b31f7 --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4640a59d454bf9f69d67460592ab71e7cdce3afa0c0a6f0cf4500bb4ac0b8381 +size 1719116358 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.yaml b/vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8960b4d1c5d8a3a7004c99de8c955f567494488b --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_instvoc_duality_v1.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: null + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.ckpt b/vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..e4042424f484dbb10d5f76f807c04f9b7f307e23 --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4a69558708f2857e36ac86a0e03ed95c4e3d8b9c5b8113963987d0d7df7e20f +size 1719116358 diff --git a/vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.yaml b/vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8960b4d1c5d8a3a7004c99de8c955f567494488b --- /dev/null +++ b/vocal/vocal_instrumental_dual/melband_roformer_instvox_duality_v2.yaml @@ -0,0 +1,51 @@ +audio: + chunk_size: 485100 + dim_f: 1024 + dim_t: 256 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 384 + depth: 6 + stereo: true + num_stems: 2 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: True + dim_freqs_in: 1025 + sample_rate: 44100 # needed for mel filter bank from librosa + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: False + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + +training: + instruments: + - Vocals + - Instrumental + target_instrument: null + use_amp: True + +inference: + batch_size: 1 + dim_t: 1101 + overlap_size: 24255 diff --git a/vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.ckpt b/vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b18c4240a94738ac82c3c48628b592ff0c06682b --- /dev/null +++ b/vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46f3f92abdd76ebe192f7c9e6167bf4d75fcffd21628f7844e7e48da42bcd7eb +size 944683368 diff --git a/vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.yaml b/vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7930da8484e0cdfd5ee3265cb09062d9b8b06df9 --- /dev/null +++ b/vocal/vocal_instrumental_dual/metal_roformer_inst_mesk_preview.yaml @@ -0,0 +1,49 @@ +audio: + chunk_size: 881559 + dim_f: 1024 + dim_t: 2000 + hop_length: 441 + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.0 +model: + dim: 384 + depth: 8 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + num_bands: 60 + dim_head: 64 + heads: 8 + attn_dropout: 0 + ff_dropout: 0 + flash_attn: true + dim_freqs_in: 1025 + sample_rate: 44100 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: false +training: + instruments: + - vocals + - other + target_instrument: other + other_fix: true + use_amp: true +inference: + batch_size: 1 + dim_t: 2000 + overlap_size: 44078 diff --git a/vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.ckpt b/vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..b4055a4b1a95d052eb780e85ebca885ce82c4de4 --- /dev/null +++ b/vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ea6e9495685045e6b4e66174131be5d19808bb0d6d1a1ba717d238f9380e8d0 +size 168124336 diff --git a/vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.yaml b/vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.yaml new file mode 100644 index 0000000000000000000000000000000000000000..62c83aeec2e39cc04c63b9bc34f8e99616315a66 --- /dev/null +++ b/vocal/vocal_instrumental_dual/model_vocals_htdemucs_sdr_8.78.yaml @@ -0,0 +1,120 @@ +audio: + chunk_size: 485100 # samplerate * segment + min_mean_abs: 0.000 + hop_length: 1024 + +training: + batch_size: 8 + gradient_accumulation_steps: 1 + grad_clip: 0 + segment: 11 + shift: 1 + samplerate: 44100 + channels: 2 + normalize: true + instruments: ['vocals', 'other'] + target_instrument: null + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + num_epochs: 1000 + num_steps: 1000 + optimizer: adam + lr: 9.0e-05 + patience: 2 + reduce_factor: 0.95 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental + +inference: + overlap_size: 242550 + batch_size: 8 + +model: htdemucs + +htdemucs: # see demucs/htdemucs.py for a detailed description + # Channels + channels: 48 + channels_time: + growth: 2 + # STFT + num_subbands: 1 + nfft: 4096 + wiener_iters: 0 + end_iters: 0 + wiener_residual: false + cac: true + # Main structure + depth: 4 + rewrite: true + # Frequency Branch + multi_freqs: [] + multi_freqs_depth: 3 + freq_emb: 0.2 + emb_scale: 10 + emb_smooth: true + # Convolutions + kernel_size: 8 + stride: 4 + time_stride: 2 + context: 1 + context_enc: 0 + # normalization + norm_starts: 4 + norm_groups: 4 + # DConv residual branch + dconv_mode: 3 + dconv_depth: 2 + dconv_comp: 8 + dconv_init: 1e-3 + # Before the Transformer + bottom_channels: 512 + # CrossTransformer + # ------ Common to all + # Regular parameters + t_layers: 5 + t_hidden_scale: 4.0 + t_heads: 8 + t_dropout: 0.0 + t_layer_scale: True + t_gelu: True + # ------------- Positional Embedding + t_emb: sin + t_max_positions: 10000 # for the scaled embedding + t_max_period: 10000.0 + t_weight_pos_embed: 1.0 + t_cape_mean_normalize: True + t_cape_augment: True + t_cape_glob_loc_scale: [5000.0, 1.0, 1.4] + t_sin_random_shift: 0 + # ------------- norm before a transformer encoder + t_norm_in: True + t_norm_in_group: False + # ------------- norm inside the encoder + t_group_norm: False + t_norm_first: True + t_norm_out: True + # ------------- optim + t_weight_decay: 0.0 + t_lr: + # ------------- sparsity + t_sparse_self_attn: False + t_sparse_cross_attn: False + t_mask_type: diag + t_mask_random_seed: 42 + t_sparse_attn_window: 400 + t_global_window: 100 + t_sparsity: 0.95 + t_auto_sparsity: False + # Cross Encoder First (False) + t_cross_first: False + # Weight init + rescale: 0.1 + diff --git a/vocal/vocal_instrumental_dual/neo_inst_beta.ckpt b/vocal/vocal_instrumental_dual/neo_inst_beta.ckpt new file mode 100644 index 0000000000000000000000000000000000000000..9af6d8f0a720d93b3ddbaee78c5677797f2e30ff --- /dev/null +++ b/vocal/vocal_instrumental_dual/neo_inst_beta.ckpt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec725df7ccea51b3250004aa0299a7fbac3dfaab2beb19f2dd04d2a4296736b0 +size 204483448 diff --git a/vocal/vocal_instrumental_dual/neo_inst_beta.yaml b/vocal/vocal_instrumental_dual/neo_inst_beta.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1da018613e379b6d742f96181d7d8e2c32072f53 --- /dev/null +++ b/vocal/vocal_instrumental_dual/neo_inst_beta.yaml @@ -0,0 +1,138 @@ +audio: + chunk_size: 882000 + dim_f: 1024 + dim_t: 1700 # don't work (use in model) + hop_length: 441 # don't work (use in model) + n_fft: 2048 + num_channels: 2 + sample_rate: 44100 + min_mean_abs: 0.000 + +model: + dim: 256 + depth: 12 + stereo: true + num_stems: 1 + time_transformer_depth: 1 + freq_transformer_depth: 1 + linear_transformer_depth: 0 + freqs_per_bands: !!python/tuple + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 2 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 4 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 12 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 24 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 48 + - 128 + - 129 + dim_head: 64 + heads: 8 + attn_dropout: 0. + ff_dropout: 0. + flash_attn: true + dim_freqs_in: 1025 + stft_n_fft: 2048 + stft_hop_length: 441 + stft_win_length: 2048 + stft_normalized: false + mask_estimator_depth: 2 + multi_stft_resolution_loss_weight: 1.0 + multi_stft_resolutions_window_sizes: !!python/tuple + - 4096 + - 2048 + - 1024 + - 512 + - 256 + multi_stft_hop_size: 147 + multi_stft_normalized: False + mlp_expansion_factor: 4 + use_torch_checkpoint: False # it allows to greatly reduce GPU memory consumption during training (not fully tested) + skip_connection: False # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training + +training: + batch_size: 2 + gradient_accumulation_steps: 1 + grad_clip: 0 + instruments: ['vocals', 'other'] + patience: 3 + reduce_factor: 0.95 + target_instrument: other + num_epochs: 1000 + num_steps: 1000 + augmentation: false # enable augmentations by audiomentations and pedalboard + augmentation_type: simple1 + use_mp3_compress: false # Deprecated + augmentation_mix: true # Mix several stems of the same type with some probability + augmentation_loudness: true # randomly change loudness of each stem + augmentation_loudness_type: 1 # Type 1 or 2 + augmentation_loudness_min: 0.5 + augmentation_loudness_max: 1.5 + q: 0.95 + coarse_loss_clip: true + ema_momentum: 0.999 + # optimizer: prodigy + optimizer: adam + # lr: 1.0 + lr: 1.0e-5 + other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental + use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true + +inference: + batch_size: 2 + dim_t: 1700 + overlap_size: 44100 + normalize: false