Nov 14 3 days ago 21:54 Render the card: working usage, repository, checkpoint and version rows Rendered by scripts/model_card.py in TigreGotico/phoonnx from the voice config. Text outside the table and the usage section is unchanged. 20:17 Point moved repo ids at Phoonnx and remove a false checkpoint note The training checkpoint is in this repository for fine-tuning. 19:39 Add the Model details table Rendered by scripts/model_card.py in TigreGotico/phoonnx from the voice config. Text outside the table is unchanged. 19:11 Name the unpublished training dataset without a link TigreGotico/tts_vc_mcv-scripted-v23.0_oc_miro is not on the Hub, so a link to it returns 404. 17:07 Set keep_punctuation: true Set keep_punctuation: true. The model was trained on text with its clause marks; with the key absent, phoonnx strips them before synthesis. ASR round trip on the pt-PT miro voice (parakeet-tdt-0.6b-v3, 12 sentences, 3 rounds): WER 21/17/20 of 92 with the flag off, 14/13/11 with it on. Oct 9 19:31 restore the training checkpoint Restores the Lightning checkpoint removed from main on 2026-09-10; copied server-side from revision 11c5f948. Sep 23 00:50 docs: the How to use snippet loads the voice through the model manager; from_hf does not exist Sep 10 11:27 note where the training checkpoint lives 11:26 remove the training checkpoint; a verified copy is at /media/hdd16/phoonnx-training-checkpoints/phoonnx_oc_miro_unicode/epoch=788-step=105726.ckpt 11:04 replace the sample: the previous one was synthesized with the default voice 10:24 add a sample of this voice Aug 10 20:39 docs: correct lineage wording (fine-tuned from closest-language base) 19:50 docs: full model card (ownership, usage, datasets) Jun 7 19:38 lang_code -> oc-FR (BCP-47) Dec 17, 2025 17:51 Create README.md 17:28 Upload 3 files 17:25 initial commit