# Sound Skill Dependencies

numpy>=1.24.0
soundfile>=0.12.0
librosa>=0.10.0
pydub>=0.25.0
torch>=2.0.0
funasr>=1.4.0
rich>=13.0.0
pyyaml>=6.0
modelscope>=1.39.0
# In-process MLX backend for VoxCPM2 / Qwen3-ASR / Qwen3-ForcedAligner (Apple Silicon).
mlx-audio>=0.4.4,<0.5.0
transformers==4.57.6
# MLX model downloads from HuggingFace (mlx-community).
huggingface_hub==0.36.2
# Speaker diarization (ModelScope 3D-Speaker pipeline): HDBSCAN clustering + modelscope
# config loader deps that are not pulled in transitively.
hdbscan>=0.8.33
simplejson>=3.19.0
sortedcontainers>=2.4.0
scikit-learn>=1.3.0

# Dev
pytest>=8.0.0
pytest-asyncio>=0.23.0
