Menu

Force-Align for Pocketsphinx

Help
2009-01-08
2014-04-16
  • Jagadeesh Balam

    Jagadeesh Balam - 2009-01-08

    How can I do a force-align to train a model for pocketsphinx. I am not able to print the noisedict symbols that are recognized with the base model. Please let me know how to do this for pocketsphinx.

    Thanks,
    J

     
    • Nickolay V. Shmyrev

      Pocketsphinx and forced alignment aren't related at all. You can enable forced alignment during training in a config file etc/sphinx_train.cfg

       
      • Nicky Wang

        Nicky Wang - 2014-04-09

        can you be more specific? I also want to know about this, thanks very much.

         
  • J. Grillparzer

    J. Grillparzer - 2014-04-16

    With bw during adaptation process (following the adaptation tutorial) you can specify -outphsegdir to output phone segmentation files. According to this thread https://sourceforge.net/p/cmusphinx/discussion/help/thread/66c6c2fd it's also possible to feed phone segmentation files to bw. Also more information about the phseg file format can be found there.

    There's a little perl script for converting phone segmentation files to text label files for audacity here: http://cmusphinx.sourceforge.net/wiki/speakerdiarization ( works with -fsize=100 ).

    Whether pocketsphinx_batch -hypseg parameter works the same way like -outphsegdir I don't know and haven't tried out yet. It would be nice if you could share what you've found out.

    These are the little pieces of information about that topic I was able to get. I guess there is more out there. But I hope it helps for a start.

     

Log in to post a comment.

Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.