ASA 124th Meeting New Orleans 1992 October

5pSP11. Recovering articulator trajectories using task dynamics and a genetic algorithm.

Richard S. McGowan

Haskins Labs., 270 Crown St., New Haven, CT 06511

A method for recovering the movement of articulators from speech acoustics is proposed. This method consists of optimizing a fit between acoustic data and the corresponding acoustic parameters produced by an articulatory synthesizer (model synthesizer) over a time interval corresponding to either a vowel with closure, or a release with a vowel. In the simulation experiments reported here, the data consists of the first three formant frequency trajectories produced by the Haskins articulatory synthesizer, ASY [Rubin et al., J. Acoust. Soc. Am. 70, 321--328 (1981)]. The articulatory synthesizer used to fit the data is identical, but the parameters that control the synthesizer and are varied to provide the best fit are task dynamic variables as specified in a gestural score [E. L. Saltzman and K. G. Munhall, Ecol. Psychol. 1, 333--382 (1989); C. P. Browman and L. Goldstein, J. Phonet. 18, 299--320 (1990)]. A genetic algorithm is used to optimize the fit between the formant data and the formants produced by the model synthesizer. This approach appears to be able to recover the basic tongue body tongue tip, and jaw movements of the ASY vocal tract, but with substantial computing effort. [Work supported by NIH Grant No. DC-02147 to Haskins Laboratories.]