Confusion between the format of the VGGish features in the TGZ and in the VGGish features created by featurefy.py

[simondurand]

The VGGish features in the TGZ are so far in the format:

{"time_points": [...], "features": [[...]]}

The VGGish features obtained with featurefy.py are so far in the format:

{time, features, features_z}

However, the "features" key in the TGZ VGGish features do not correspond to the to the "features" key in the featurefy.py VGGish features, but to its "features_z" key.

This might create confusion when trying to apply to trained model to new audio.

[simondurand]

@ejhumphrey I suggest two possible solutions:

1. We modify the keys in either the TGZ or the featurefy.py function so that the same things are named the same way.
2. We use a different name in the TGZ that can't be confused with what is produced by featurefy.py

I prefer 1) as I think it is easier for someone jumping in the project who wants to train a model with the TGZ data and then apply it to new audio files with featurefy.py to deal with the same key for the same things.
I would even suggest that the output of featurefy.py to be {time_points, features}, with "time_points" corresponding to what is currently "time" and "features" corresponding to what is currently "features_z". We probably do not need 3 fields. And this way, we only have 1-2 line of code change.