I was wondering, does CMU plan to release a support for PNC files as it does
for standard MFC files?
I have produced an PNCC acoustic model for French and the results showed an
improvement of about 12% WER against its MFCC counterpart. For that, I used
the matlab script produced buy CMU, embedded into a perl application that
allowed to automatize the pnc file creation.
But the process is slow and only experimental.
Thank you
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
12% absolute WER reduction sounds quite impressive ...
My understanding from looking at some papers is that they made the comparisons
using 16KHz speech.
Does anyone knows whether the relative advantages and robustness of PNCC are
preserved with 8KHz telephone-grade speech?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello,
I was wondering, does CMU plan to release a support for PNC files as it does
for standard MFC files?
I have produced an PNCC acoustic model for French and the results showed an
improvement of about 12% WER against its MFCC counterpart. For that, I used
the matlab script produced buy CMU, embedded into a perl application that
allowed to automatize the pnc file creation.
But the process is slow and only experimental.
Thank you
Hi Remus
The C code that existed for PNCC was rather dirty so I wouldn't rely on it
being released.
I understand. But still, after reading some papers and conducting this study
(which involved more than 5000 audio files), it seems PNCC is promising.
Hope a pncc "wave2feat" to be ultimately released :)
Best regards,
Hi,
12% absolute WER reduction sounds quite impressive ...
My understanding from looking at some papers is that they made the comparisons
using 16KHz speech.
Does anyone knows whether the relative advantages and robustness of PNCC are
preserved with 8KHz telephone-grade speech?