Dynamic Articulatory Model of Speech Production

Abstract

A dynamic articulatory model of speech production is described. From a discrete phonemic input, it generates a continuous description of the configuration of the articulatory mechanism in a midsagittal section. The model is divided into a “state” and “operators,” which act as agents for modifying the state by trying to manipulate it (i.e., responding articulatory structures) toward abstract “goals.” This bifurcation allows many of the constraints of the real system to be included in the model in a natural manner. By this means, the contextual effects of connected speech can be generated by less ad hoc methods than usually required by other schemes of speech modeling and/or synthesis. The present model concerns itself only with vowels and stops, but it is felt that the general methodologies are applicable to all speech sounds. A motion-picture film of model action, as simulated on a large time-shared computer system, is shown. [Work supported in part by the U. S. Air Force Cambridge Research Laboratories, Office of Aerospace Research, and by Project MAC, and MIT research program sponsored by the U. S. Department of Defense.]