Menu ▾ ▴

Problem in segmenting a file

Help
Pankaj
2011-01-04
2012-09-22

  • Pankaj

    Pankaj - 2011-01-04

    Hello,

    I have been using the cont_fileseg.c given in sphinxbase for segmenting a long
    speech file into a no of utterances.
    But I have got a file, which doesn't gets splitted by the pocketsphinx.
    Pocketsphinx considers it as just one single utterance. There are sufficient
    silent zones between any two utterances in the file.

    Link for the file : http://www.mediafire.com/file/79zj5872c8qnhm1/amit_long_1
    6khz_4Jan2011.raw

    Please help.

    Regards

     

  • Nickolay V. Shmyrev

    Nickolay V. Shmyrev - 2011-01-11

    Sorry, this is a bug in ad code that requires a fix. No time to fix it right
    now unfortunately.

     

  • Pankaj

    Pankaj - 2011-01-12

    Hello Nshmyrev,

    Can you give me a hint. I will try to fix the code and also share the code if
    successful. I have gone through the ad code extensively and a hint may
    definitely be of help

    Regards

     

  • Pankaj

    Pankaj - 2011-01-14

    Hello,
    My further observations are that the file splitting of the previously
    mentioned file happens properly if I do any of the following:

    1. Even if a very small portion (minimum 0.002 s) from the beginning of the file is deleted.
      OR
    2. If the value of CONT_AD_CALIB_FRAMES is changed from the default (CONT_AD_POWHISTSIZE * 2) to anything else say (CONT_AD_POWHISTSIZE*3)
      OR
    3. If the value of CONT_AD_MIN_NOISE is changed from the default 2 to 4.
      OR
    4. If the rewinding of the input file after calibration is not done, i.e. if the rewind(fp) statement in the file cont_fileseg.c is commented.
     

Log in to post a comment.