Investigating the consonant-vowel boundary: Perceptual contributions to sentence intelligibility

Abstract

Speech sounds divide into two broad categories: vowels and consonants. Vowels have a distinct perceptual advantage over consonants in determining sentence intelligibility [Kewley‐Port, Burkle, and Lee, J. Acoust. Soc. Am. (in press)]. However, specifying the segmentation of consonants and vowels is problematic because of their extensive acoustic overlap due to coarticulation. The current study used TIMIT sentences to investigate perceptual contributions of consonants and vowels across systematic changes in the consonant‐vowel (C‐V) boundary. Sentences were presented to listeners with either consonant or vowel information replaced by noise. The C‐V boundary was shifted by six specific proportions of the vowel, such that consonant duration increased while vowel duration decreased, yielding 12 different noise replacement conditions. The percent of words listeners repeated correctly was scored. A two‐to‐one vowel advantage for intelligibility at the traditional C‐V boundary was confirmed. Initial results from ten listeners suggest that functions of the shifted C‐V boundaries for consonants and vowels differentially contribute to intelligibility. Preliminary comparison of these functions suggests that both consonants and vowels contribute equally (50%) to intelligibility at about a 20% proportional boundary shift. Results will be interpreted in terms of phonological versus acoustic accounts of speech perception. (Supported by NIH)