Chapter 35. Speech Communication Academic and Research Staff Graduate Students 1. Constraints and Strategies in Speech Production Formulation and Implementation of a Model for Lexical Access in Running Speech 2.1 Progress in Model Development 3. Acoustic/articulatory Studies of Speech-sound Producti

Introduction The objective of this research is to refine and test a theoretical framework in which words in the lexicon are represented as sequences of segments and syllables and these units are represented as complexes of auditory/acoustic and somatosensory goals. The motor programming to produce sequences of sensory goals utilizes an internal neural model of relations between articulatory motor commands and their acoustic and somatosensory consequences. The relations between those articulatory motor commands and the movements they generate are influenced by biomechanical constraints, which include characteristics of individual speakers' anatomies and more general dynamical properties of the production mechanism. To produce an intelligible sound sequence while accounting for biomechanical constraints, speech movements are planned so that sufficient perceptual contrast is achieved with minimal effort. There are individual differences in planning movements toward sensory goals that may be due to relations between production and perception mechanisms in individual speakers. In a current project, the internal model is implemented as a neurocomputational model that is used to control a vocal-tract model (an articulatory synthesizer). The combined models provide the bases of hypotheses about the planning of speech movements. To test these hypotheses, we are conducting experiments with speakers and listeners in which we measure articulatory movements, speech acoustics, perception, and brain activation. We are manipulating speech condition, phonemic context and speech sound category and we introduce transient and sustained perturbations. We are also performing modeling and simulation experiments, in which we adapt the vocal-tract model to the morphologies of individual speakers. We are testing properties of the neurocomputational model by using it to control the individualized vocal tract models in efforts to replicate those speakers' production data. During this last year, two new studies were begun and ongoing studies have been continued or completed. 1.1 Development of facilities Software has been acquired and used to generate natural-sounding synthetic speech stimuli for a battery of perceptual tests that will be used to measure speakers' auditory acuity. Scripts have been developed to facilitate the use of the synthesis software for the rapid generation of large numbers of stimuli in continua for vowel, stop and sibilant contrasts. Software has been developed to record and evaluate responses in labeling and discrimination tasks. Preliminary testing of these paradigms is underway. 1.2 Sensorimotor adaptation in the production of vowels We have used a DSP board and its control software to modify the first formant frequency (F1) of vowels …