Takes a text file of filenames and makes a list of filenames

Loads wavdata into torch array

Flatten dense tensors into a contiguous 1D buffer. Assume tensors are of same dense type. Since inputs are dense, the resulting tensor will be

View a flat buffer using the sizes of tensors. Assume that tensors are of same dense type, and that flat is given by _flatten_dense_tensors. A

Modifies existing model to do gradient allreduce, but doesn't change class so you don't need "module"

forward_input[0] = mel_spectrogram: batch x n_mel_channels x frames forward_input[1] = audio: batch x time