Production and perception of English vowels

Abstract

This study examined the relation between production of nine English vowels and perception of the synthesized vowels by 14 American male and female speakers. Fant’s bandwidth equations were employed to dynamically tune to the varied formant frequency values. A remarkable r2 value was obtained from the regressional analysis between the center formant frequency values at which the subjects perceived the same vowel quality from the discriminatory test and the formant values of the synthesis models. Males and females perceived the synthesized vowels in the same way with converging center formant values and similar ranges of the same vowel quality. There was a strong link between production and perception of male and female speakers. The average r2 value was very high, which suggests a very lawful relation between production and perception. From the individual analyses we found that listeners adjusted the criteria for vowel discrimination in relation to their own vocal tracts. This result suggests that speaker normalization can be pursued by finding an individual regression equation between the reference and observed formant patterns. Also, the present data proved that human listeners possess a very accurate perceptual mechanism to extract invariant features from complex acoustic stimuli.