4th International Conference on Spoken Language Processing
Philadelphia, PA, USA
Several studies on place perception of prevocalic stop consonants have shown that the apparent perceptual weight of release burst and formant transitions depends on the vowel context: bursts carry higher perceptual weight in high front vowel contexts like /i/ than in low non-front vowel contexts like /a/, while the reverse holds for formant transitions. This finding is generally interpreted as reflecting a context-dependent "reweighting" of burst and transition cues by the perceptual system. In this paper it is shown that the observed effect can be entirely accounted for by contextual variation of the distributions of the relevant cues themselves: Naturally produced stop bursts appear to be acoustically more distinctive in high front vowel contexts than in low non-front vowel contexts, while the reverse is true for formant transitions. The apparently context-dependent perceptual weight of burst and transitions can be reproduced with such cue distributions, even if the classification model itself contains fixed, context-independent linear boundaries. This claim is supported with acoustical and perceptual data of a burst-splicing experiment involving burst-spliced stop-vowel utterances containing the stops /p, t, k/ and vowels /a, i, y, u/.
Bibliographic reference. Smits, Roel (1996): "Context-dependent relevance of burst and transitions for perceived place in stops: it's in production, not perception", In ICSLP-1996, 2470-2473.