# Use the Python environment that runs ComfyUI.
# PyTorch / TorchAudio / TorchVision and NumPy are supplied by ComfyUI.
# Keep the existing CUDA builds; no PyTorch version pins here.
transformers>=4.52.0,<5
qwen-omni-utils>=0.0.9
# Required by the Qwen/Librosa audio import chain; declare explicitly.
audioread>=3.0.1
omegaconf>=2.3.0
torchdiffeq
x_transformers>=1.31.14
safetensors
accelerate>=0.33.0
PyYAML>=6.0
silero-vad>=6.0
# Optional text normalization: pynini cannot build with MSVC.
# AuK already falls back to the source text when this module is unavailable.
WeTextProcessing>=1.2.0; sys_platform != "win32"
pyloudnorm>=0.2.0
openai>=1.0.0
tencentcloud-sdk-python-asr>=3.0.0
funasr>=1.3,<2
modelscope>=1.20,<2
soundfile>=0.12
