Write function to take audio files to spectrograms

[iKevinY]

Input

A dictionary of speaker names mapped to lists of paths to audio files:

{
    'speaker_1': ['file_1a.wav', 'file_1b.wav'],
    'speaker_2': ['file_2a.wav'],
}

Output

X_train, y_train, X_validation, y_validation, consisting of shuffled data from the provided audio files (rows of X are our spectrograms and y are the labels). In addition, we want a list y_mapping which is a list of speakers in the order of the classes in y (ie. if y_mapping is ['speaker_1', 'speaker_2'], then (0, 1) would correspond to predicting speaker_2).

[iKevinY]

This exists as part of #107.