Acoustic properties underlying vowel intelligibility in words with mixed consonantal contexts
Sarah Hargus Ferguson, Diane Kewley-Port, Aaron P. Brown · The Journal of the Acoustical Society of America · 2001
This study explored the relative importance of spectral target, dynamic formant, and duration information to vowel identification in mixed consonantal contexts. Vowel intelligibility was varied naturally by one talker producing test words in sentences using clear and conversational speech. In /bVd/ context only, vowels identified by normal-hearing listeners in noise were significantly more intelligible (15%) in clear speech than in conversational speech. Clear and conversational tokens of the 10 vowels also differed significantly on several spectral and temporal measures. Regression analyses revealed that intelligibility differences for individual vowels were well predicted by various combinations of spectral target, dynamic formant, and duration metrics [S. H. Ferguson and D. Kewley-Port, J. Acoust. Soc. Am. 106, 2272 (1999)]. A second data set was collected concurrently with the /bVd/ data, with the same listeners identifying vowels in mixed CVC contexts. A significant intelligibility advantage (18%) for clear speech was found for these materials. However, regression analyses were less straightforward for vowels in CVC context than in /bVd/ context. Several acoustic metrics were investigated to determine whether intelligibility differences could be modeled for vowels in mixed consonantal contexts. The results will be discussed in terms of theories of vowel perception. [Work supported by NIHDCD-02229, the American Speech-Language-Hearing Foundation, and NIHDCD-00012.]