Menu

Problem in adapting an acoustic model

Help
2011-03-29
2012-09-22
  • Tasbeer Ahsan

    Tasbeer Ahsan - 2011-03-29

    Hi,

    I am trying a to make an application for Android that will recognize speech
    commands in another language (non-English)

    For this purpose I have downloaded and tried the PocketsphinxAndroidDemo and
    ran it successfully for the models provided in hub4wsj_sc_8k folder in
    pocketsphinx.

    As the next step of my application I developed the language models, phone
    file, dictionaries etc for the language and trained using SphinxTrain which
    generated some model_parameters.

    What I am trying to do is replace the models used in the demo with my models
    from .ci_cont folder generated during training using SphinxTrain and follow
    this http://cmusphinx.sourceforge.net/wiki/tutorialadapt tutorial to create my models.
    however when I try this step :

    ./bw \
    -hmmdir hub4wsj_sc_8k \
    -moddeffn hub4wsj_sc_8k/mdef.txt \
    -ts2cbfn .cont. \
    -feat 1s_c_d_dd \
    -svspec 0-12/13-25/26-38 \
    -cmn current \
    -agc none \
    -dictfn Trainme.dic \
    -ctlfn Trainme.fileids \
    -lsnfn Trainme.transcription \
    -accumdir .

    I get the following error

    FATAL_ERROR: "mod_inv.c", line 354: # of features in mixw file, 1, is
    inconsistent w/ prior setting, 3

    I am using a continuous model hence I changed the " -ts2cbfn .semi. \" to "
    -ts2cbfn .cont. \" which should have solved this problem but is not working.

    Can anyone please help me out or tell me if my steps of adapting the model by
    replacing the existing ones is correct.

     
  • Rafael Oliveira

    Rafael Oliveira - 2011-03-29

    Hi,

    Recently i did exactly what you are trying to do now, but, using a own
    database and this problem does not showed up. I have a guess you could try. In
    the same tutorial that you are using like baseline there is a comment about
    the svspec parameter. It is said that:

    For example, for most continuous models (like the ones used by Sphinx4) you
    don't need to include the svspec option

    So, how i was adapting a continuous model, i have not used the svspec, and,it
    worked.

    Note that i have not tried to execute the bw program using svspec, so i can
    not affirm that the problem is happening because of it.

     

Log in to post a comment.