Menu

Converting HTK acoustic models to Sphinx

Help
Anonymous
2011-08-04
2012-09-22
  • Anonymous

    Anonymous - 2011-08-04

    Hi,

    I'm trying to convert HTK acoustic models to Sphinx format using the Python
    HTK Converter from https://cmusphinx.svn.sourceforge.net/svnroot/cmusphinx/tr
    unk/htk2s3conv/

    I'm working primarily on Windows 7, where I have been able to run the
    converter after installing Python and Python PLY. To test, I'm trying to
    convert htk_wsj_all_8000_2 from http://www.keithv.com/software/htk/us/

    I have converted this to ASCII format as the script was not working on the
    binary form. After converting to ASCII, the script gave a message requesting
    the removal of "sp" models as they are not supported by Sphinx, which I did.

    Now when I run the script, I get the following message and a list of biphones:

    "Not all HMMs could be classified as either monophone or triphone"

    And there is no output. I have tried the following output formats:

    convert.py hmmdefs tiedlist output. (to get output as a prefix)
    convert.py hmmdefs tiedlist c:\output\ (to get the output in a folder named
    output)
    convert.py hmmdefs tiedlist ~\test\ (to get the output in a folder)

    None of these seem to be working. I am not sure if there is the problem with
    the syntax I'm using to define the output or if the script is just not able to
    convert for these particular models.

    Any suggestions would be much appreciated.

    Thanks.

     
  • Nickolay V. Shmyrev

    "Not all HMMs could be classified as either monophone or triphone"

    Biphones are not supported in CMUSphinx just as sp. You need to remove them
    from your model before conversion.

     
  • Anonymous

    Anonymous - 2011-08-04

    Thank you very much for the prompt response, I'll remove the biphones and
    report on the results shortly.

     
  • Anonymous

    Anonymous - 2011-08-08

    I removed the biphones, but got further errors of missing phone models that
    were included in triphones. So, I am now going to try converting Sphinx models
    to HTK format (my aim was to compare the two). Many thanks for the help.

     

Log in to post a comment.