Code to calculate MFCC features for TED-LIUM using python_speech_features and format them for the net about deepspeech

[kdavis-mozilla]

There is a Python package python_speech_features that can be used to calculate MFCC features.

[kdavis-mozilla]

The issue6 branch[1] has example code showing how data can be integrated.

[kdavis-mozilla]

@andrenatal I've made a WIP commit f1d697a addressing this issue. It's not finished but the majority of the code is there. The largest pain was having to write the queue and threading code by hand.

As a warning, do no try to use the standard TensorFlow Queues and Readers[1]. They will not work.

As we require paddings, we would have to use a PaddingFIFOQueue. However, a PaddingFIFOQueue doesn't allow the other thing we require SparseTensors, as TensorFlow's Queues do not work with SparseTensors[2]. So this basically means we have to write the code by hand as I did.

[andrenatal]

Thank you @kdavis-mozilla! I'll jump on it right now!

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.