Refactored most of the code so that it supports stft_layer.to(device) like other torch modules now, where stft_layer = nnAudio.Spectrogram.STFT() or any other spectrogram class available in nnAudio.
Since this version no longer uses the device argument, it is not compatible with the older versions. Adjustments are required to upgrade your code from older versions to 0.2.0.
For example, nnAudio.Spectrogram.STFT(device=device) in the older version should be rewritten as nnAudio.Spectogram.STFT().to(device) in 0.2.0.
If you are using nnAudio inside your PyTorch model, then model.to(device) is enough to transfer nnAudio layers to the device you want.