Compensatory Articulation During Speech: Evidence from the Analysis and Synthesis of Vocal-Tract Shapes Using an Articulatory Model

Abstract

Temporal variations observed in the vocal-tract profiles during continuous speech were studied using a factor analysis. Roughly 1000 frames of cineradiographic and labiofilm data corresponding to 10 French sentences uttered by two speakers have been analyzed. The factor analysis permits us to describe the observed profiles as the sum of a small number of linear components. Since these components can be interpreted by articu- latory terms, the linear model is considered as an articulatory one. For example, the tongue profiles can be specified by the following four parameters; the mandibular and tongue-dorsal positions, the dorsal shape, and the apex position. The entire vocal-tract configurations including the frontal lip-opening shapes can be described with reasonable accuracy by as few as seven parameters in total. With this model the temporal variations are described in terms of the frame-by-frame samples of these articulatory parameters. We observe that "target" parameter values for the same vowel vary significantly presumably due to different phonetic contexts. An acoustic calculation with the model predicts that a particular pair of articulators can compensate acoustically each other. For example, by an appropriate adjustment of the tongue-dorsal position, the model is capable of producing the same F1-F2 pattern for different jaw position, or vice-versa. The compensation between the jaw and the dorsal positions, however, is possible only for unrounded vowels. In the case of rounded vowels, the jaw position can be compensated by the lip aperture. The measured "target" values of the paired parameters indicated a linear relationship, suggesting that the speakers actually exploit the inter-articulator compensation in the speech production. This explains the observed large "target" value variability. The comparison of parameter trajectories for the same sentences uttered by the two speakers indicates similitude rather than difference, suggesting that the manner of the production involving the com- pensatory articulation could be relatively invariant. 1.