Menu

error, when run perl ./scripts_pl/50.cd_hmm_

Help
2012-02-04
2012-09-22
  • nguyen duy nam

    nguyen duy nam - 2012-02-04

    hi, my proplem when run perl ./scripts_pl/50.cd_hmm_tied/slave_convg.pl
    and i veiwe in logdir 50.cd_hmm_tied/words.1.1-1.bw
    "model_def_io.c", line 436: Unable to open
    /Application/Data/speechtotext/demo/mothai/model_architecture/words.1000.mdef
    for reading; No such file or directory
    i dont know why i miss file words.1000.mdef
    pls, help me

     
  • Nickolay V. Shmyrev

    Please check logs for details. Not just the last log, but previous logs too.
    That's the common rule for training, see troubleshooting section in

    http://cmusphinx.sourceforge.net/wiki/tutorialam

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-04

    thanks, i passed it, and i trained ok 100% with 39 audio file each file length
    7s, but when i run pocketsphinx_continuous
    it incorrectly, say the word program return other word.
    can you give me an advice
    can you tell me when use words.cd_cont_1000, and when to use
    words.cd_cont_1000_2 words.cd_cont_1000_1 ... Because when I finished it
    generated in the folder model_parameters:

    words.cd_cont_1000
    words.cd_cont_1000_1
    words.cd_cont_1000_2
    words.cd_cont_1000_4
    words.cd_cont_1000_8
    words.cd_cont_initial
    words.cd_cont_untied
    words.ci_cont
    words.ci_cont_flatinitial

    thanks

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-04

    you can test it for me, my folder trainning:
    http://www.mediafire.com/?9lip6lgx59u7f57

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-05

    hi Nickolay
    can you test and for me an advice.
    and can you to me know how to train one word with many( ex 10) speaker 10
    times. do i understand that to do it i need setup file fileids:

    speaker_1/file_1
    speaker_2/file_2
    speaker_3/file_3
    speaker_4/file_4
    speaker_5/file_5
    speaker_6/file_6
    speaker_7/file_7
    speaker_8/file_8
    speaker_9/file_9
    speaker_10/file_10

    and file transcription

    ONE file_1
    ONE file_2
    ONE file_3
    ONE file_4
    ONE file_5
    OnE file_6
    ONE file_7
    ONE file_8
    ONE file_9
    ONE file_10

    and file dic have only one word "ONE O N E". sorry if i wrong, you can fix it
    for me.
    thanks

     
  • Nickolay V. Shmyrev

    Sorry, in your database there are so many things done wrong that I really
    suspect you missed the tutorial

    http://cmusphinx.sourceforge.net/wiki/tutorialam

    If you go there and read the tutorial text carefully you could get it done
    right. Now it's all plain wrong.

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-05

    thanks your reply, yes in tutorial http://cmusphinx.sourceforge.net/wiki/tuto
    rialam
    have many thing i
    not clearly. i will read again to make clear them.

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-05

    hi, nickolay.
    if my database audio haven't audio test, so i don't need file test.fileids and
    test.transcript?

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-07

    hi Nickolay
    i readed again the tutorial text, and i thing i make them ok, but when i run
    pocketsphinx_continuous -hmm model_parameters/words.cd_cont_untied -lm
    etc/words.lm.DMP -dict etc/words.dic
    it till return result wrong. can you test again for me.
    my folder train:
    http://www.mediafire.com/?d1d1m838z9xlkql
    i record audio in Audacity( 1 Mono, 16 000 HZ, 16 bit PCM) and save file
    Microsoft WAV 16 bit

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-07

    i dont know why it error 100%
    **

    ./scripts_pl/decode/slave.pl
    MODULE: DECODE Decoding using models previously trained
    Decoding 2 segments starting at 0 (part 1 of 1)
    0%
    Aligning results to find error rate
    SENTENCE ERROR: 100.0% (2/2) WORD ERROR RATE: 100.0% (2/2)
    **

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-07

    pls. help me

     
  • Nickolay V. Shmyrev

    Well now it's better but you missed the first paragraph of the tutorial:

    When you need to train

    You want to create an acoustic model for new language/dialect
    OR you need specialized model for small vocabulary application
    AND you have plenty of data to train on:
    1 hour of recording for command and control for single speaker
    5 hour of recordings of 200 speakers for command and control for many speakers
    10 hours of recordings for single speaker dictation
    50 hours of recordings of 200 speakers for many speakers dictation
    AND you have knowledge on phonetic structure of the language
    AND you have time to train the model and optimize parameters (1 month)

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-09

    ok, thanks you very much.
    i knowed about it. and the folder train that i sended to you do test my setup
    right or wrong. then now i'm recording database.
    thanks

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-09

    hi. nickolay
    1. can you tell me when use words.cd_cont_1000, and when to use words.cd_cont_1000_2 words.cd_cont_1000_1 ... Because when I finished it generated in the folder model_parameters:

    words.cd_cont_1000
    words.cd_cont_1000_1
    words.cd_cont_1000_2
    words.cd_cont_1000_4
    words.cd_cont_1000_8
    words.cd_cont_initial
    words.cd_cont_untied
    words.ci_cont
    words.ci_cont_flatinitial

    1. other question: how to decrease SENTENCE ERROR and WORD ERROR RATE
      thanks
     
  • Nickolay V. Shmyrev

    can you tell me when use words.cd_cont_1000, and when to use
    words.cd_cont_1000_2 words.cd_cont_1000_1 ... Because when I finished it
    generated in the folder model_parameters:

    See http://cmusphinx.sourceforge.net/wiki/tutorialam#using_the_model

    other question: how to decrease SENTENCE ERROR and WORD ERROR RATE

    Train better model

     
  • nguyen duy nam

    nguyen duy nam - 2012-02-10

    thanks.

    Train better model

    What is a parameter to setting better model?

     
  • Nickolay V. Shmyrev

    What is a parameter to setting better model?

    Parameters are described in tutorial. Model becomes better mainly because of
    bigger and more carefully annotated data, not because of a parameters.

     

Log in to post a comment.