Contrast and normalization in vowel perception

Abstract

The work reported in this paper is an attempt to understand vowel normalization better by investigating the relationship between vowel normalization and vowel contrast. In the first experiment, vowels from a “hood”–“hud” continuum were presented at two levels of fundamental frequency (F0) using two types of presentation. In one condition, tokens were blocked by F0. In the other, tokens with different F0 levels were randomly intermixed with each other (as in the typical F0 normalization experiment). In the mixed presentation, subjects identified the high F0 items most often as “hood” and the low F0 items most often as “hud”. In the blocked condition, there was no reliable difference between the high and low F0 continua. This pattern of results suggests that a contrast effect is at work. Therefore, four models of perceptual contrast were tested in simulations using auditorily-based spectra produced by a model which incorporates two levels of processing, (1) narrow-band auditory filtering (R. D. Patterson, J. Acoust. Soc. Am. 1976, 59, 640) and (2) wide-band integration (L. A. Chistovich, J. Acoust. Soc. Am. 1985, 77, 789). The experiment’s results could be approximated by either of two models: an auditory figure/ground model, and a talker contrast model. A second experiment distinguished between these two models. The auditory figure/ground model predicts that in a cross-series anchoring experiment (in which tokens with high F0 are used to anchor the low F0 continuum and tokens with low F0 are used to anchor the high F0 continuum) the boundary of the vowel identification function will be shifted toward the vowel quality of the anchoring stimulus. The talker contrast model predicts that the vowel quality of the anchoring stimulus is less important than its F0 and that the phoneme boundary will be shifted in the same direction regardless of the vowel quality of the anchoring stimulus. The results of the experiment quite unambiguously supported the predictions of the talker contrast model.