I was having the same error message (MFCC read of arctic_0001..mfc failed).
Beware the BW and map_adapt appear to create an unaltered model when using
arctic20.listoffiles (as in the tutorialadapt wiki) but when you switch to
arctic20.fileids thats when I began to see the error you guys are getting:
Then I tried what yaksea mentioned and added:
-cepext mfc
To eliminate the error message, and it appears to work. I couldn't figure out
how to edit the source code like NancyXII mentioned, so I didn't attempt that.
I will let you know if my model actually improves its accuracy based on the
adaptation corpus.
thanks,
TJ Miler
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I finally got it to work, then played around with Tau parameters from 0.001 to
100. For tuning to myself on the an4 trigram batch test, a tau parameter of 1
gave me 98% accuracy versus the standard WSJ model which gave me ~96%
accuracy. Looks like it works
thanks,
TJ Miler
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Guys,
I was having the same error message (MFCC read of arctic_0001..mfc failed).
Beware the BW and map_adapt appear to create an unaltered model when using
arctic20.listoffiles (as in the tutorialadapt wiki) but when you switch to
arctic20.fileids thats when I began to see the error you guys are getting:
Then I tried what yaksea mentioned and added:
-cepext mfc
To eliminate the error message, and it appears to work. I couldn't figure out
how to edit the source code like NancyXII mentioned, so I didn't attempt that.
I will let you know if my model actually improves its accuracy based on the
adaptation corpus.
thanks,
TJ Miler
All:
I finally got it to work, then played around with Tau parameters from 0.001 to
100. For tuning to myself on the an4 trigram batch test, a tau parameter of 1
gave me 98% accuracy versus the standard WSJ model which gave me ~96%
accuracy. Looks like it works
thanks,
TJ Miler