A physiologically- based computer model will be developed and used to test hypotheses about biomechanical constraints on speech motor control. The model has two components: a biomechanical model of the vocal tract and a control system. The biomechanical model will have realistic dynamical and anatomical properties and its outputs will be articulatory displacements, the shape of the vocal tract (from which airflows, pressures and noise sources will be calculated), the vocal-tract transfer function and sound. To help achieve realistic dynamics, a non- linear tissue description is used in finite element models of the tongue and the lips, based on a large deformation representation. The model will be developed in the context of a theoretical framework, according to which speech movements are programmed to achieve sequences of goals that are characterized in articulatory and acoustic terms. The design of the controller is influenced by observations about speech production; it will be a modular hierarchical system in which neural networks are used as building blocks to represent information about the controller's dynamics. The complete controller will take sequences of acoustic and articulatory goals as inputs and learn to generate commands that result in acceptable patterns of time-varying vocal-tract shapes and a sound output, so the model can be used to investigate the constraining effects of its properties on speech motor control strategies. The biomechanical model will be customized to the morphology of one or two individual speakers with the use of anatomical (MRI) data, and some EMG data from the speakers will be used to help validate the model's properties. The model's development and testing will be guided by having it emulate kinematic, aerodynamic, force and acoustic data from the same speakers. (The data from speakers are being obtained in another project in the laboratory). Since these emulations will involve a model of the vocal tract with realistic dynamical and anatomical properties, hypotheses about the control may be more representative of the actual processes than has been the case previously.
