vllm[audio]<0.31,>=0.30.0
soundfile>=0.12
huggingface-hub

[bench]
jiwer>=3
zhon
openai

[test]
pytest>=8
