Menu

About content of model file mdef

Help
Willy
2018-11-28
2019-05-06
  • Willy

    Willy - 2018-11-28

    Hi, I've used the pretrained model in pocketsphinx/model/en-us/en-us to do KWS, which is proveided by the package itself, and it worked very well in my application.
    For the interest of model content, I open the 'mdef' file by using command 'pocketsphinx_mdef_convert', and it showed me these information

    0.3
    42 n_base
    137053 n_tri
    548380 n_state_map
    5126 n_tied_state
    126 n_tied_ci_state
    42 n_tied_tmat

    which should mean there are 5126 senones in the acoustic model.
    However when I open 'means' and 'variances' files (by printp) they showed only the means and variances of states of 42 monophones, not 5126 senones.
    So I want to know where are the Gaussian parameters of those remaining senones.
    Also, I'm cusrious about what database was used to train this model.
    Thanks.

     
    • Nickolay V. Shmyrev

      The model is PTM model, so the gaussians are shared across senones of the same core phone. Only mixture weights differ.

      The database for the model is not public.

       

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.