Convert Training Observations to Feature Vectors

[chadlagore]

TODO

• This is an bit of a detour, but it may speed up our training.
• In our raw training set, observations are just raw audio binary (hundreds of "feature" columns).
• The goal is to transform the raw binary into feature vectors (python_speech_features).
• This should reduce the number of columns by an order of magnitude, and for some models, improve accuracy.

Use the collector module!