Menu

Sphinx Train giving error

Help
2011-12-20
2012-09-22
  • vijayabharadwaj gsr

    I am doing Sphinx training. It starts fine. Every test passed in initial
    steps. Trees have been built. In the Module 50, first gaussian iterations are
    finished without problem. when it splits to two gaussians I am gettting error.

    MODULE: 50 Training Context dependent models (2011-12-18 03:17)

    Phase 1: Cleaning up directories:
    accumulator... logs... qmanager... completed

    Phase 2: Copy CI to CD initialize

    init_mixw Log File

    Current Overall Likelihood Per Frame = 5.27179472944965

    Convergence Ratio = 0.280849058955256

    Baum welch starting for 2 Gaussian(s), iteration: 4 (1 of 1)

    bw Log File
    completed

    Only 0 parts of 1 of Baum Welch were successfully completed

    Parts 1 failed to run!

    When I opened Log File, I found the following error


    -0.16
    bw: gauden.c:1348: gauden_scale_densities_bwd: Assertion `finite(den)' failed.
    column defns
    <seq>
    <id>
    <n_frame_in>
    <n_frame_del>
    <n_state_shmm>
    <avg_states_alpha>
    <avg_states_beta>
    <avg_states_reest>
    <avg_posterior_prune>
    <frame_log_lik>
    <utt_log_lik>
    ... timing info ...
    utt> 0 10-1 2964 0 936 14 Sun Dec 18 03:28:03 2011 </utt_log_lik></frame_log_lik></avg_posterior_prune></avg_states_reest></avg_states_beta></avg_states_alpha></n_state_shmm></n_frame_del></n_frame_in></id></seq>

    Can you please guide me what could be the reason

     
  • Nickolay V. Shmyrev

    When you report about the training problem please learn to provide as much
    information as possible:

    What sphinxtrain version are you using

    What is the size of the training database

    What modifications in default training parameters have you made

    Such information could help to answer your problem quickly. For example one
    could easily figure out that you just don't have enough data to train the
    parameters you configured.

     
  • vijayabharadwaj gsr

    I am sorry for not providing enough information. I am using sphinxtrain build
    nightly which I downloaded in July. I have enough speech data which is 25
    hours. I am training for 16 gaussians and 1000 senones.

     
  • vijayabharadwaj gsr

    I have already trained successfully 20 hours data with 16 gaussians. The
    vocabulary is about 7000. But Now I have added 4 hours more and I am training
    it again. Vocabulary increased to 13000. I don't see the question of not
    having enough data for training.

     

Log in to post a comment.