wavenet-waveform-coder / evaluation_metrics.json
wq2012's picture
Update retrained model and metrics (v0.2.2)
d6ea89d verified
Raw History Blame Contribute Delete
7.47 kB
{
"num_eval_utterances": 16,
"section_3_2_waveform_coding_rates": {
"mean_entropy_bits_per_sample": 6.0642,
"mean_entropy_kbps": 97.03,
"cross_entropy_bits_per_sample": 5.5285,
"cross_entropy_nll_nats_per_sample": 3.832,
"cross_entropy_rate_kbps": 88.46,
"wavenet_parametric_w_wo_cross_entropy_bits_per_sample": 5.4827,
"wavenet_parametric_w_wo_cross_entropy_rate_kbps": 87.72,
"wavenet_parametric_w_w_cross_entropy_bits_per_sample": 5.456,
"wavenet_parametric_w_w_cross_entropy_rate_kbps": 87.3,
"arithmetic_coded_bits_per_sample": 5.523,
"arithmetic_coded_rate_kbps": 88.37,
"lossless_exact_match_rate": 1.0
},
"section_3_3_codec_comparison": {
"Reference (16 kHz)": {
"bitrate_kbps": 256.0,
"pesq_wb_mos_lqo": 4.644,
"stoi": 1.0,
"f0_rmse_hz": 0.0,
"objective_mos_lqo": 4.644,
"mcd_db": 0.0,
"lsd_db": 0.0,
"segmental_snr_db": 45.0,
"high_band_energy_ratio_percent": 5.378
},
"8-bit mu-law (128 kb/s)": {
"bitrate_kbps": 128.0,
"pesq_wb_mos_lqo": 4.141,
"stoi": 0.999,
"f0_rmse_hz": 0.103,
"objective_mos_lqo": 4.141,
"mcd_db": 0.47,
"lsd_db": 0.87,
"segmental_snr_db": 35.144,
"high_band_energy_ratio_percent": 5.388
},
"WaveNet Waveform (WW)": {
"bitrate_kbps": 88.455,
"pesq_wb_mos_lqo": 4.141,
"stoi": 0.999,
"f0_rmse_hz": 0.103,
"objective_mos_lqo": 4.141,
"mcd_db": 0.47,
"lsd_db": 0.87,
"segmental_snr_db": 35.144,
"high_band_energy_ratio_percent": 5.388
},
"Hybrid Mode-Switching (WW + W_w)": {
"bitrate_kbps": 5.427,
"pesq_wb_mos_lqo": 1.456,
"stoi": 0.818,
"f0_rmse_hz": 16.288,
"objective_mos_lqo": 1.456,
"mcd_db": 6.382,
"lsd_db": 6.758,
"segmental_snr_db": -1.103,
"high_band_energy_ratio_percent": 1.048
},
"AMR-WB (23.05 kb/s)": {
"bitrate_kbps": 23.05,
"pesq_wb_mos_lqo": 4.006,
"stoi": 0.988,
"f0_rmse_hz": 7.494,
"objective_mos_lqo": 4.006,
"mcd_db": 1.523,
"lsd_db": 3.218,
"segmental_snr_db": 5.711,
"high_band_energy_ratio_percent": 2.492
},
"WaveNet Parametric (W_w, 2.4 kb/s)": {
"bitrate_kbps": 2.4,
"pesq_wb_mos_lqo": 1.507,
"stoi": 0.817,
"f0_rmse_hz": 15.958,
"objective_mos_lqo": 1.507,
"mcd_db": 6.677,
"lsd_db": 7.064,
"segmental_snr_db": -2.112,
"high_band_energy_ratio_percent": 0.09
},
"WaveNet Parametric (W_wo, 2.4 kb/s)": {
"bitrate_kbps": 2.4,
"pesq_wb_mos_lqo": 1.56,
"stoi": 0.814,
"f0_rmse_hz": 15.412,
"objective_mos_lqo": 1.56,
"mcd_db": 7.256,
"lsd_db": 7.484,
"segmental_snr_db": -2.077,
"high_band_energy_ratio_percent": 0.052
},
"MELP (2.4 kb/s)": {
"bitrate_kbps": 2.4,
"pesq_wb_mos_lqo": 1.708,
"stoi": 0.807,
"f0_rmse_hz": 17.825,
"objective_mos_lqo": 1.708,
"mcd_db": 8.091,
"lsd_db": 7.989,
"segmental_snr_db": -2.759,
"high_band_energy_ratio_percent": 0.021
},
"Codec 2 (2.4 kb/s)": {
"bitrate_kbps": 2.4,
"pesq_wb_mos_lqo": 1.548,
"stoi": 0.793,
"f0_rmse_hz": 13.002,
"objective_mos_lqo": 1.548,
"mcd_db": 8.383,
"lsd_db": 8.212,
"segmental_snr_db": -2.187,
"high_band_energy_ratio_percent": 0.018
},
"Speex (2.15 kb/s)": {
"bitrate_kbps": 2.15,
"pesq_wb_mos_lqo": 1.477,
"stoi": 0.788,
"f0_rmse_hz": 40.598,
"objective_mos_lqo": 1.477,
"mcd_db": 8.055,
"lsd_db": 7.971,
"segmental_snr_db": -1.242,
"high_band_energy_ratio_percent": 0.024
}
},
"section_3_4_speaker_identification": {
"mulaw_speech_eer": {
"eer_percent": 2.6785714285714284,
"eer_ci95_low_percent": 0.8928571428571428,
"eer_ci95_high_percent": 10.725446428571432,
"ci_95_low_percent": 0.8928571428571428,
"ci_95_high_percent": 10.725446428571432,
"threshold": 0.3827601362123674,
"mean_genuine_cosine": 0.5310029472095787,
"mean_impostor_cosine": -0.022036171423084377
},
"wavenet_parametric_w_w_eer": {
"eer_percent": 12.5,
"eer_ci95_low_percent": 3.125,
"eer_ci95_high_percent": 37.5,
"ci_95_low_percent": 3.125,
"ci_95_high_percent": 37.5,
"threshold": 0.4416057608377243,
"mean_genuine_cosine": 0.5472275705238807,
"mean_impostor_cosine": 0.14966936472793405
},
"wavenet_parametric_w_wo_eer_mulaw_verifier": {
"eer_percent": 13.392857142857142,
"eer_ci95_low_percent": 4.464285714285714,
"eer_ci95_high_percent": 33.50446428571429,
"ci_95_low_percent": 4.464285714285714,
"ci_95_high_percent": 33.50446428571429,
"threshold": 0.42581775881416317,
"mean_genuine_cosine": 0.5683047755218508,
"mean_impostor_cosine": 0.18131676375533382
},
"wavenet_parametric_w_wo_eer": {
"eer_percent": 12.5,
"eer_ci95_low_percent": 3.5491071428571432,
"eer_ci95_high_percent": 32.16517857142857,
"ci_95_low_percent": 3.5491071428571432,
"ci_95_high_percent": 32.16517857142857,
"threshold": 0.34334123158536545,
"mean_genuine_cosine": 0.5337333474836207,
"mean_impostor_cosine": -0.06123304633513006
},
"codec2_2400_eer": {
"eer_percent": 25.0,
"eer_ci95_low_percent": 12.5,
"eer_ci95_high_percent": 50.44642857142857,
"ci_95_low_percent": 12.5,
"ci_95_high_percent": 50.44642857142857,
"threshold": 0.6822124058642703,
"mean_genuine_cosine": 0.7492089943548031,
"mean_impostor_cosine": 0.5129870441804585
},
"melp_2400_eer": {
"eer_percent": 25.0,
"eer_ci95_low_percent": 12.488839285714286,
"eer_ci95_high_percent": 38.47098214285715,
"ci_95_low_percent": 12.488839285714286,
"ci_95_high_percent": 38.47098214285715,
"threshold": 0.6201746179382821,
"mean_genuine_cosine": 0.7565372295600474,
"mean_impostor_cosine": 0.48238106747532267
},
"speex_2150_eer": {
"eer_percent": 36.160714285714285,
"eer_ci95_low_percent": 12.5,
"eer_ci95_high_percent": 50.44642857142857,
"ci_95_low_percent": 12.5,
"ci_95_high_percent": 50.44642857142857,
"threshold": 0.6862327789197117,
"mean_genuine_cosine": 0.740649492401773,
"mean_impostor_cosine": 0.5453444186954882
},
"amr_wb_23050_eer": {
"eer_percent": 12.5,
"eer_ci95_low_percent": 2.6785714285714284,
"eer_ci95_high_percent": 21.875,
"ci_95_low_percent": 2.6785714285714284,
"ci_95_high_percent": 21.875,
"threshold": 0.3985332753516293,
"mean_genuine_cosine": 0.5707722609885149,
"mean_impostor_cosine": 0.08427913132809407
},
"triangle_test_w_w_vs_w_wo": {
"mean_cosine_orig_vs_w_wo": 0.6313405586986083,
"mean_cosine_orig_vs_w_w": 0.6884149530360879,
"mean_cosine_w_wo_vs_w_w": 0.9037546800111116,
"mean_cosine_advantage": 0.057074394337479684,
"embedding_odd_one_out_identification_percent": 93.75,
"triangle_test_identification_percent": 93.75,
"triangle_accuracy_percent": 93.75,
"abx_preference_percent": 93.75,
"abx_preference_w_w_over_w_wo_percent": 93.75,
"random_chance_baseline_percent": 33.33
}
}
}