| # duplex server / offline driver (same env as the Dynamic Accent Flash TTS: torch, CosyVoice deps) | |
| fastapi | |
| uvicorn[standard] | |
| websockets | |
| httpx | |
| silero-vad # MIT | |
| fasttext-wheel # GlotLID (cis-lmu/glotlid, Apache-2.0) for language ID | |
| huggingface_hub | |
| soundfile | |
| librosa | |
| numpy | |
| # ASR worker env (separate venv is fine): omnilingual-asr>=0.2.0 (Apache-2.0), huggingface_hub, fastapi, uvicorn (Dynamic ASR weights: Modularcomputing/dynamic-asr) | |
| # LLM server: vllm | |