torch>=1.10.0
transformers>=4.20.0
datasets
numpy
soundfile
librosa