pytorch_vggish use python_speech_features to get MFCC Features. use torchaudio for your dataset.py. use my network.py.