Abstract

The acoustic-phonetic properties of speech sounds vary substantially across languages, and across talkers within a single language. While many factors contribute to this variation (e.g., sociolinguistic, physiological), there is considerable evidence that acoustic-phonetic variation is highly structured among speech sounds. Structured variation of this type can be observed in the covariation among vowels in the F1xF2 vowel space, as talkers form relatively congruent, but shifted vowel spaces (e.g., Joos, 1948; Nearey, 1978), as well as in the covariation of talker mean voice onset time among aspirated stop consonants (e.g., Chodroff & Wilson, 2017). The presence of covariation reveals constraints on permissible variation in language- and talker-specific sound systems (e.g., Liljencrants & Lindblom, 1972), and may be exploited in generalized perceptual adaptation to novel talkers.

In this talk, I will examine the predictions of two constraints on phonetic implementation that could give rise to covariation across talkers. The constraints are formalized as target uniformity, which requires similar (or identical) phonetic realization of a distinctive feature value within a talker, and contrast uniformity, which requires a comparable phonetic difference in sounds that contrast in a feature across talkers. Evidence from several case studies of sibilant fricatives implicates target uniformity, but not contrast uniformity as a constraint on phonetic realization.

I will also discuss the implications of phonetic covariation for perceptual adaptation to novel talkers. Covariation indicates mutual predictability among speech sounds, such that evidence from one speech sound could be used to infer properties of a second speech sound. I present evidence that perceptual knowledge of phonetic covariation is consistent with many patterns of generalized adaptation, but general auditory factors (e.g., spectral contrast) may have primary influence on short-term adaptation.