a spec augment layer as a module in PyTorch model almost all calculation step is in tensor, can be used in cuda. just use as a layer as dropout, content: time_mask, freq_mask
Seems that a little accumulation needed, wellcome anyone to contribute. sample output is in pic.png
