scipy
fire
sklearn
s3prl
torchaudio==0.9.0
sentencepiece
