Menu ▾ ▴

Partial hypothesis problem

Help
Matt Hall
2011-03-25
2012-09-22
<< < 1 2 (Page 2 of 2)

  • Anuj Kumar

    Anuj Kumar - 2011-07-03

    My scenario is i have maximum 20-30 words. and i expect a sentence in its
    exact same order. so can say , expect one word at a time. So do you have any
    advice to increase the accuracy by changing the parameters or number of hmms
    or method of searching?

    Change the dictionary to only have those 20-30 words, and generate a new
    language model from the sentences that you expect. Also, instead of a
    statistical LM, you could also use a rule-based LM, though I don't know if
    pocketSphinx specifically supports that.

     

  • Anonymous

    Anonymous - 2011-07-03

    Youy can use JSGF grammars by passing the -jsgf command line arg. See the
    cmdln_macro.h source file. I use the the W3C docs for grammar docs
    http://www.w3.org/TR/jsgf/

     

  • sarinsukumar

    sarinsukumar - 2011-07-15

    Hi all
    thanks for the links and advices, especially to Anuj

    it seems like the pocket sphinx is doing a great recognition. But i am facing
    a new problem now. The recognition accuracy for female sounds and children
    sounds are not great with respect to the accuracy for male voices, is there
    special HMMs available for women and children voice?

    Thanks in advance for the replays.

     

  • Nickolay V. Shmyrev

    Nickolay V. Shmyrev - 2011-07-15

    is there special HMMs available for women and children voice?

    You can adapt existing acoustic model to improve accuracy. See

    http://cmusphinx.sourceforge.net/wiki/tutorialadapt

     
<< < 1 2 (Page 2 of 2)

Log in to post a comment.