gradio
librosa
loguru
numpy<2
pyannote.audio
torch
huggingface_hub[hf_transfer]