• DocumentCode
    1692761
  • Title

    Contributions of the high-RMS-level segments to the intelligibility of mandarin sentences

  • Author

    Fei Chen ; Wong, Lena L. N.

  • Author_Institution
    Div. of Speech & Hearing Sci., Univ. of Hong Kong, Hong Kong, China
  • fYear
    2013
  • Firstpage
    7810
  • Lastpage
    7814
  • Abstract
    Recent evidence suggests that segments carrying more spectral changes [e.g., consonant-vowel boundaries in the middle root-mean-square (RMS) level segments] are important to predict the intelligibility of English sentences. Nevertheless, considering the difference between Mandarin and English languages, it is hypothesized that the high-RMS-level segments might provide more perceptual information to the intelligibility of Mandarin speech. Two studies were conducted in this paper to assess the relative contributions of the high-RMS-level segments to the intelligibility of Mandarin sentences, i.e., speech perception and intelligibility prediction. Results show that 1) Mandarin sentences containing the high-RMS-level (i.e., above the overall RMS level of the whole utterance) segments are more intelligible (i.e., recognition rate up to 91%) than those with the middle-RMS-level segments; and 2) the high-RMS-level segments, which carry more vowel and tonal information, contribute more in predicting the intelligibility of Mandarin sentences in noise.
  • Keywords
    mean square error methods; natural language processing; speech intelligibility; speech processing; English sentences; Mandarin sentences; Mandarin speech; high-RMS-level segments; intelligibility prediction; root-mean-square; spectral changes; speech perception; Auditory system; Correlation coefficient; Indexes; Noise; Speech; Speech recognition; System-on-chip; Speech perception; intelligibility prediction;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on
  • Conference_Location
    Vancouver, BC
  • ISSN
    1520-6149
  • Type

    conf

  • DOI
    10.1109/ICASSP.2013.6639184
  • Filename
    6639184