FORMANT TRANSITIONS : TEASING APART CONSONANT AND VOWEL CONTRIBUTIONS

Acoustic data and models of consonant releases suggest that many CV transitions can be regarded as a sequence of two components: (1) an initial local change due to the release of the primary consonant articulator and (2) slower changes of F2 and F3 as the tongue body and jaw move away from positions as supporting articulators for the consonant constriction to their positions as primary articulators for the vowels. INTRODUCTION As is known, the transitions of the formants in consonant-vowel syllables reflect the identity of the articulator that makes the consonantal constriction, and where this articulator is placed. The transitions are also influenced by the articulatory configuration for the following vowel, and how much the vowel configuration is anticipated during the consonant constriction. (e.g., [1]. 121. 131. [41. [51, [6]). The release of a consonant in a symmetric VCV can be viewed as having two components, particularly when the primary articulator is the lips or tongue blade. The first component is the initial local movement of the primary articulator. The second, slower component consists of the movements of the tongue body and/or jaw, and possibly rounding of the lips, toward positions required for the vowel. Here we explore the contributions of both components to formant transitions, and focus on labials and alveolars. GENERAL METHOD We are addressing this problem in two ways: (1) by calculation of formant movements for vocal tract models that are manipulated to change in a stepwise manner from a consonant-like to a vowel-like configuration; (2) by examination of natural speech. The use of modeling techniques augments the analysis of natural speech in several important ways. First, in natural speech, when the area of the constriction is still quite small, the sound may be dominated by a transient or frication burst created at the constriction, making it difficult to determine the natural frequencies of the vocal tract as a whole during this initial part of the release. Second, the initial formant movements can be quite rapid, and measurement is subject to the wellknown problem of time-frequency tradeoffs in accuracy. This problem is sidestepped with modeling which calculates formant frequencies for each step of the movement. Third, at present it is not possible to completely determine the vocal tract shape from the acoustic signal. Modeling allows one to be quite explicit about the vocal tract shape.