Menu

Problem in acostic model training in punajbi language

Help
puneet
2015-12-03
2015-12-04
  • puneet

    puneet - 2015-12-03

    Hello all
    I am new in speech recognition field. i am trying to build acoustic model in punjabi language for 5 commands. When i try to train it, gives error
    ERROR: "gauden.c", line 1682:
    Variance (mgau= 8, feat= 0, density=0, component=1) is less then 0. Most
    probably the number of senones is too high for such a small training
    database. Use smaller $CFG_N_TIED_STATES.
    please help
    I have reduced the $CFG_N_TIED_STATES=1
    then also same error is shown.

     
  • Asad Ullah

    Asad Ullah - 2015-12-03

    please re-read this wiki, may be you will find your solution.
    http://cmusphinx.sourceforge.net/wiki/tutorialam

     
    • puneet

      puneet - 2015-12-03

      I have prepared all files and seems all files are correct
      .dic
      .phone.
      .filler
      .fileids
      .transcription
      I hae read all instructions from the link you have provided but then also I dont know whether problem is in my files or somewhere else

       
      • Nickolay V. Shmyrev

        The problem is that you don't have enough data for training specific phoneme. You can check which phoneme is that in mdef file in model_architecture folder. You need to look for state number 8.

        To get more precise help on the issue you need to share a model training folder.

         

Log in to post a comment.