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
Integrates HumeAI's TADA (Text-Acoustic Dual Alignment) speech-language
model as a new TTS engine. TADA uses a novel 1:1 token-audio alignment
that produces coherent speech over long sequences (700s+).
Two model variants:
- tada-1b: English-only, ~4GB, built on Llama 3.2 1B
- tada-3b-ml: 10 languages, ~8GB, built on Llama 3.2 3B
Backend uses the Encoder for voice prompt encoding with caching, and
TadaForCausalLM with flow-matching diffusion for generation. Supports
bf16 inference on CUDA, forces CPU on macOS (MPS compatibility).
Installed with --no-deps due to torch>=2.7 pin conflict; descript-audio-codec
and torchaudio added as explicit sub-dependencies.
James Pine