Menu

ST: Assertion n == feat->stream_len[0...

Help
marekl
2010-09-06
2012-09-22
  • marekl

    marekl - 2010-09-06

    Hi Nicolay,
    I have started tests with MMIE training getting in the MODULE: 60 Lattice
    Generation, Phase2 the following error

    INFO: info.c(65): Host: 'ubuntu'
    INFO: info.c(69): Directory: '/home/monika/workspace/training'
    INFO: info.c(73): /home/monika/workspace/training/bin/.libs/lt-sphinx3_align Compiled on: Aug 25 2010, AT: 16:49:24
    
    INFO: cmd_ln.c(512): Parsing command line:
    /home/monika/workspace/training/bin/.libs/lt-sphinx3_align \
        -hmm /home/monika/workspace/training/model_parameters/boolware.cd_cont_8000 \
        -senmgau .cont. \
        -mixwfloor 1e-08 \
        -varfloor 0.0001 \
        -dict numlat/boolware.falign.dict \
        -fdict numlat/boolware.falign.fdict \
        -ctl /home/monika/workspace/training/etc/boolware.fileids \
        -ctloffset 11628 \
        -ctlcount 3876 \
        -cepdir /home/monika/workspace/training/feat \
        -cepext .mfc \
        -insent numlat/boolware.aligninput \
        -outsent numlat/boolware.alignedtranscripts.4 \
        -wdsegdir numlat,CTL \
        -beam 1e-60 \
        -agc none \
        -cmn current \
        -varnorm no \
        -feat 1s_c \
        -ceplen 13
    
    (...)
    
    INFO:   Initialization of the log add table
    INFO:   Log-Add table size = 29356 x 2 >> 0
    INFO:   
    INFO: feat.c(848): Initializing feature stream to type: '1s_c', ceplen=13, CMN='current', VARNORM='no', AGC='none'
    INFO: cmn.c(142): mean[0]= 12.00, mean[1..12]= 0.0
    INFO:   Reading Feature Space Transform from: /home/monika/workspace/training/model_parameters/boolware.cd_cont_8000/feature_transform
    lt-sphinx3_align: lda.c:128: feat_read_lda: Assertion `n == feat->stream_len[0]' failed.
    

    parameters for LDA and MMLT training are the same as in previous trainings
    MMIE parameters:

    $CFG_NUMLAT_DIR = "numlat";
    $CFG_DENLAT_DIR = "denlat";
    $CFG_LANGUAGEMODEL = "boolware.DMP";
    $CFG_LANGUAGEWEIGHT = "11.5";
    $CFG_BEAMWIDTH      = "1e-100";
    $CFG_WORDBEAM       = "1e-80";
    $CFG_WORDPENALTY    = "0.2";
    
    $CFG_ABEAM = "1e-50";
    $CFG_NBEAM = "1e-10";
    $CFG_PRUNED_DENLAT_DIR = "pruned_denlat";
    
    $CFG_LATTICE_DIR = "lattice";
    
    $CFG_MMIE_TYPE = "best";
    $CFG_MMIE_CONSTE = "3.0";
    $CFG_MMIE_MAX_ITERATIONS = 5;
    

    Do you have any idea what could cause such a problem?

     
  • marekl

    marekl - 2010-09-06

    just crossed my mind that this might by the same problem with strncpy in
    Sphinx3 as in Sphinx Train. Is it possible?

     
  • Nickolay V. Shmyrev

    Hello Marek

    Sphinx3_align doens't support cepwin features out-of-box yet. You need to
    apply cepwin patch to sphinxbase in order to read cepwin features.

     
  • marekl

    marekl - 2010-09-06

    Thanks a lot, by the way - it would be useful to add such remark to MMIE
    tutorial for Sphinx4 users (now I can remember that you have mentioned
    cepwin.diff file on your blog, but till now I really didn't needed this patch
    to train and work with sphinx4)

     
  • Nickolay V. Shmyrev

    It would be actually useful to add cepwin support to sphinxbase, probably it
    could be named like "1s_3x" and "1s_4x" unfortunately I'm unable to find the
    time to do that.

     
  • Nickolay V. Shmyrev

    K, I did that. With sphinxbase trunk you can use feature type 1s_3c and 1s_4c
    for cepwin 3 and 4.

     
  • marekl

    marekl - 2010-09-10

    Great! Thx

     

Log in to post a comment.