Abstract:

Dynamic models in speech production have been traditionally based on linear regression analysis of articulatory data. The purpose of this work is to introduce a stochastic calculus methodology to the analysis of such data in speech production models. The issue of variability encountered in the analysis and description of articulatory movements during speech may be addressed with the implementation of a stochastic term, thus placing the variability in a controlled probabilistic framework. Upper lip, lower lip, jaw, and velum movements tracked optoelectronically at Haskins Laboratories [H. B. Kollia, V. Gracco, and K. Harris, J. Acoust. Soc. Am. 98, 1313--1327 (1995)] during production of a test utterance are described in this manner. The authors believe that this algorithm has wide applicability since it yields confidence interval rather than point value estimates for kinematic data description. [Work supported by NSF.]