Created by: reyoung

Users want to use deep speech 2 model for transfer learning. Users want to extract the speech features by DS2 model, just like they extract image features by VGGNet.

And then, many interesting things could be done, for example.

• We can compare the similarity of two speakers. The difference will show how clear the input speech was, thus allowing us to determine how clear or unclear one's pronunciation of a phoneme may be.