search

mpezeshki / CTC-Connectionist-Temporal-Classification

Theano implementation of CTC.
Apache License 2.0
74 stars 26 forks source link

On data formats and the way to ge them for dataLoader.py #7

Closed konts6102 closed 8 years ago

konts6102 commented 8 years ago

First of all, I am impressed this exciting source and appreciate all contributors.

Question 1. Is it possible for kaldi scripts to create a set of input files with which ctc-stanford training script can run?

I think that all input files for an execution are: key#.txt, feat.bin and alis#.txt. Examples of these files are in the URL: http://deeplearning.stanford.edu/lexfree/timit/

Question 2. If the answer of the previous question is NO, then how can I create these three kinds of files from my wav files and transcripts of them?

I know methods and scripts to extract the mfcc and log mel filter bank features as described in the URL https://github.com/jameslyons/python_speech_features

I think that a set of feature vectors of a wav file can be an input for dataLoarder.py. But I am not sure how the rest of three files can be obtained.