Closed razor1179 closed 7 years ago
yes you would use a different wspecifier (see Kaldi IO mechanisms) that specifies to output to an uncompressed file. An example of a wspecifier that writes to an archive (ark) in text format is: "ark,t:/tmp/new_features.ark". The compressed ark format is gzip so you could use gunzip to uncompress. You could also use the "copy-feats" program to convert formats. Here's an example of doing the opposite direction (text to compressed binary format)
Hi,
I am currently trying to train the RNN using Tensorflow, but with the features extracted from EESEN. I would like to know if there is any change I can make to the codes used to obtain the features such that the .ark files created are not compressed?
Regards, Deepak