• DocumentCode
    2280317
  • Title

    Comparison between a measure based on mutual information and frequency importance functions for consonant recognition

  • Author

    Kasturi, Kalyan S.

  • Author_Institution
    Dept. of Electron. & Commun. Eng., RMD Eng. Coll., Chennai, India
  • fYear
    2010
  • fDate
    15-17 Dec. 2010
  • Firstpage
    471
  • Lastpage
    474
  • Abstract
    In this work, we seek to find how the information about consonant recognition pertaining to the various phonemes is distributed in frequency. For this task, we used the mutual information between phonetic labels and spectral energy. The analysis was performed in the perspective of cochlear implants and how information is distributed among the various channels of the cochlear implant. The calculation of mutual information was performed for each of the six channels. The computed values of mutual information between phonetic labels and spectral energy were relatively high for channels 4, 5 and 6. This indicates that those high frequency channels contain relatively high amount of information about phonetic classification for consonant stimuli and is consistent with the existing results from the intelligibility tests. Remarkable correspondence between the mutual information values of individual channels for consonant stimuli and the frequency importance functions for consonant recognition was observed.
  • Keywords
    pattern classification; spectral analysis; speech intelligibility; speech processing; consonant recognition; frequency importance functions; intelligibility tests; phonetic classification; phonetic labels; spectral energy; Frequency measurement; Indexes; Joints; Mutual information; Quantization; Speech; Speech processing; consonant recognition; frequency importance functions; mutual information; speech intelligibility;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Signal and Image Processing (ICSIP), 2010 International Conference on
  • Conference_Location
    Chennai
  • Print_ISBN
    978-1-4244-8595-6
  • Type

    conf

  • DOI
    10.1109/ICSIP.2010.5697520
  • Filename
    5697520