Kinematic factors underlying timing in connected speech

Abstract

Current understanding of the kinematic factors underlying speech timing has been largely derived from short, simple speech tasks such as citation words. However, timing analyses based upon acoustic segment durations have shown that timing rules differ for citation words and connected speech. Thus it is not clear how well temporal correlates of citation-style speech movement might generalize to connected speech. A better understanding of the articulatory bases of timing patterns associated with connected speech may aid in the refinement of natural-sounding articulatory-based speech synthesis. This study sought to determine temporal correlates of articulator fleshpoint movements in orally read speech recorded from 20 normal adult speakers of American English. Position histories associated with the tongue blade, tongue dorsum, jaw, and lower lip were segmented into series of movement strokes, and measures of distance, maximum speed, and duration were determined for each stroke. Canonical correlation analysis showed that for all articulators, these three measures were highly interrelated, and that 80% of the variance in movement duration could be accounted for by movement distance and maximum movement speed. These results suggest the possibility of predicting articulator timing based on a small number of kinematic measures. [Work supported by NIH Grant Nos. DC00820, DC03723, and DC03659.]