gradio>=5.0,<6.0
onnxruntime>=1.16
tokenizers>=0.15
huggingface_hub>=0.33
numpy>=1.24,<2.0
audioop-lts; python_version >= "3.13"