Couple of general questions:
-Are PNCC features going to replace MFCCs in future speech recognition systems ?
-Why are they(PNCC based features) gaining prominence only now ?
-For replacing MFCCs to PNCCs in Pocketsphinx, does only the front-end change or
it needs changes in the back-end too ?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
1) not necessarily
2) not that much time passed from their introduction
3) Results will be better if the one will use model trained with noice cancellation with "noise-remove" enabled.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Couple of general questions:
-Are PNCC features going to replace MFCCs in future speech recognition systems ?
-Why are they(PNCC based features) gaining prominence only now ?
-For replacing MFCCs to PNCCs in Pocketsphinx, does only the front-end change or
it needs changes in the back-end too ?
1) not necessarily
2) not that much time passed from their introduction
3) Results will be better if the one will use model trained with noice cancellation with "noise-remove" enabled.
Any reasons for still considering MFCCs in newer systems when PNCCs claim to provide significantly better noise robustness ?
True, but PNCC technique also implies usage of power law (instead of log) and gammatone filters (instead of triangular). It may be checked if that is reasonable. Please check this post: http://nshmyrev.blogspot.com/2013/06/around-noise-robust-pncc-features.html