Menu ▾ ▴

Poor Recognition in multi user whole word mod

Help
anupam
2010-12-29
2012-09-22

  • anupam

    anupam - 2010-12-29

    Hi all,
    let me give a brief description of the experiment.

    _AIM :- TO RECOGNIZE INDIAN PLACES NAME ACCURATELY. _

    _DESCRIPTION:- _

    already implemented the same experiment with single user whole word model with
    90% accuracy.

    NAME OF THE EXPERIMENT Single user whole word model Multi user whole word
    model

    NUMBER OF PLACES (words) 18 40

    NUMBER OF PERSONS VOICE RECORDED 1 16

    TOTAL TIME OF RECORDING 20 minutes 1 hour

    LANGUAGE MODEL wordlist wordlist

    ANY DIFFERENCES AMONG THE WAY
    SPEAKER UTTER THE WORD? NO YES

    EXPERIMENTED PARAMETERS IN sphix_train.cfg CFG_FINAL_NUM_DESITIES SAME
    and CFG_N_TIED_STATE

    BESTDECODING RESULT SER=7.8%,WER=7.8% SER=32.7% WER=38.6%

    RECOGNIOTION THROUGH SPHINX-4 APP 88.88% (16/18) 7.5%(3/40)

    i am putting my training folder in the following link
    http://www.mediafire.com/file/e6gzx1hlx5p6g2c/bus.jar

    please tell me where i m going wrong. If any more details are required please
    let me know.
    Thank you

    regards
    Anupam

     

  • anupam

    anupam - 2010-12-29

    hi i am sorry for the bad format of the table above please get the table from
    following link

    http://www.mediafire.com/?jktc8koqf4doo

     

  • Anupam Saha

    Anupam Saha - 2010-12-29

    Hey can we be in touch?.

    mail id: vmv_hyphen@yahoo.com

    Im working n pocketsphinx and made sample voice recognition application for
    win32, and now working on winCe. Im from India.

     

Log in to post a comment.