- Model filter in ModelManagement was an allowlist missing cosyvoice prefix,
replaced with inverse filter (everything except whisper = voice model)
- Shim training-only modules (cosyvoice.dataset.processor, matcha.utils.*)
to prevent hyperpyyaml from importing pyarrow, pyworld, lightning etc.
- Patch torchaudio.load for 2.9+ compat (torchcodec now required by default)
- Add matplotlib to requirements (matcha hifigan imports it at module level)
Integrate Alibaba's CosyVoice2-0.5B and Fun-CosyVoice3-0.5B as a new
TTS engine supporting 9 languages, zero-shot voice cloning, and instruct
control (emotions, speed, volume, dialects).
The CosyVoice source is cloned at setup time into backend/vendors/ since
no PyPI package exists. A modelscope→HuggingFace shim redirects model
downloads to the public HF repos, and a lightweight pylogger shim avoids
pulling in pytorch-lightning as a transitive dependency.
Backend: cosyvoice_backend.py, __init__.py registry, models.py regex
Frontend: engine selector, language map, Zod schema, model descriptions
Infra: requirements.txt, justfile, release.yml, Dockerfile, PyInstaller
Jamie Pine