On the role of formant transitions in vowel recognition.

Abstract

An inventory of speechlike sounds was synthesized displaying systematic variations of the rate and direction of formant transitions. These sounds were specified by a set of vowel formant patterns selected along a continuum varying from [U] to [I] they were assigned to isolated, steady-state vowels, and to the points of zero rate of formant frequency change in symmetrical consonant-vowel-consonant syllables. The time variations of formant frequencies were made convex and concave by the choice of two consonantal frames: [w-w] and [j-j]. The results obtained in a series of vowel identification experiments indicate that a listener's categorization of the continuum varied as a function of the environment and the duration of the vowel. These findings suggest that, in the recognition of monosyllabic nonsense speech, the identity of a vowel is determined not solely by the formant-frequency pattern at the point of closest approach to target, but also by the direction and rate of adjacent formant transitions. In general, subjects adjusted their categorizations of the continuum in the consonantal contexts in such a way that complete transitions between loci and vowel target were not necessary: the transitions were permitted to undershoot the target frequencies for the vowel. In particular, the excursions of formants in the [w-w] syllables tended to be overestimated. Thus, there was a tendency for the categorizations to be made so as to compensate for the formant-frequency undershoot associated with vowel reduction [B. Lindblom, “Spectrographic Study of Vowel Reduction,” J. Acoust. Soc. Am. 35, 1773–1781 (1963)]. The effects observed are discussed in terms of an active model of vowel recognition, peripheral auditory analysis, distinctive features, and previously reported observations on vowel perception.