During my experiment I have noticed that at decoding Kaldi produces in exp/mono0a/decodingdir/scoring the files 9.tra, 10.tra, 11.tra, which include what was recognized along the validation files. I have searched along the exp directory and I didn't find a file that is similar to .tra files but with additional time frames. In other words can Kaldi produce a file like HTK's recout.mlf, where you can have information about starting-time, ending time and the recognized word.
All I want is to compare my results by 10ms frame and not in the way that WER metric works.
Thanks in advance!
K.T
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
During my experiment I have noticed that at decoding Kaldi produces in
exp/mono0a/decodingdir/scoring the files 9.tra, 10.tra, 11.tra, which
include what was recognized along the validation files. I have searched
along the exp directory and I didn't find a file that is similar to .tra
files but with additional time frames. In other words can Kaldi produce a
file like HTK's recout.mlf, where you can have information about
starting-time, ending time and the recognized word.
All I want is to compare my results by 10ms frame and not in the way that
WER metric works.
Good morning everyone,
During my experiment I have noticed that at decoding Kaldi produces in exp/mono0a/decodingdir/scoring the files 9.tra, 10.tra, 11.tra, which include what was recognized along the validation files. I have searched along the exp directory and I didn't find a file that is similar to .tra files but with additional time frames. In other words can Kaldi produce a file like HTK's recout.mlf, where you can have information about starting-time, ending time and the recognized word.
All I want is to compare my results by 10ms frame and not in the way that WER metric works.
Thanks in advance!
K.T
Look at steps/get_ctm.sh
Dan
On Sun, Jul 12, 2015 at 3:11 AM, Konstantinos Themelis
kothemel@users.sf.net wrote: