Menu ▾ ▴

advanced front end for sphinx

Speech Recognition Theory
ramz
2011-09-14
2012-09-22

  • ramz

    ramz - 2011-09-14

    Hi,
    I have a basic question in speech recognition, uptill date MFCC is the most
    advanced frontend used probably in any speech recognition engine. But is there
    any other advanced frontend technique which is better than MFCC ? only a
    speech expert can answer this question, i am a beginner so i am asking here.
    Experts please help.
    Thanks in advance

     

  • Joseph S. Wisniewski

    Joseph S. Wisniewski - 2011-09-14

    There are several.
    PLP (perceptual linear prediction) and IMELDA
    Any real cepstral based front end is higher performance than what speech rec folk mistakenly call "cepstral".

    Why do you ask? The big problem is that you need acoustic models using a
    particular feature set, and there's a lot more MFCC stuff floating around
    there than other formats.

     

  • ramz

    ramz - 2011-09-15

    thanks for the reply wiz, I have trained and decoded 50 words in sphinxTrain
    and pocketsphinx, But the recognition accuracy is really low. So i thought of
    coding some advanced technique in frontend which would increase the
    recognition accuracy.
    I have read about PLP and LPC, but MFCC is more advanced than others.

    so is there any other frontend technique which is much more advanced than
    MFCC???

    thanks in advance

     

Log in to post a comment.