torch
numpy

# Audio processing
librosa
soundfile

# Model
s3prl
pyannote.audio
funasr
transformers

# WER
jiwer==3.1.0

# Normalization
zhconv
zhon