Computer simulation of the vocal tract in speech production

Abstract

Neuro-musculoskeletal function can be well characterized for most human movements, such as walking, through a combination of experimental measurements, including motion capture, electromyography, and force sensing. Movements of the vocal tract, however, pose a considerable challenge in terms of holistically measuring and visualizing the neuro-musculoskeletal processes underlying speech production. This is due to the inaccessibility of the vocal tract, the large number of small, deep muscles involved, the fast speed of movements, and the inherent three-dimensional nature of vocal tract shape during speech production. An array of measurement and imaging modalities have been tailored to vocal tract measurement, such as MRI, EMA, EPG, and ultrasound, each providing valuable, but incomplete information regarding vocal tract movements. Computer simulation of the vocal tract can play an important role in complementing these sparse experimental measurements by aiding in the fusion of multiple imaging modalities, helping to describe and visualize the 3D structures of interest, filling in the gaps in experimental measurements (both spatial and temporal), and extrapolating from the small sample sizes commonly found in speech production studies. We will discuss two types of computer simulations that are emerging as important complementary forms of speech production investigation: forward biomechanical simulation of the vocal tract articulators and probabilistic simulation of neuro-musculoskeletal parameters.