torch
torchaudio
transformers>=4.30.0
gradio>=4.20.0
numpy
accelerate
soundfile
librosa
einops
huggingface_hub