A biomechanical model of the human tongue for understanding speech production and other lingual behaviors

Adam Baker
Department of Linguistics
University of Arizona

Dissertation Abstract

A biomechanical model of the human tongue was constructed, based upon a detailed anatomical study of an actual cadaver. Data from the Visible Human Project were segmented to create a volumetric representation of the tongue and its constituent muscles. The volumetric representation was converted to a smooth NURBS-bounded solid model—for compatibility with meshing algorithms—by lofting between splines, the vertices of which were defined by the coordinates of a smoothed triangular mesh representation. Using a hyperelastic constitutive model that allowed for the addition of active stress, the model deforms in response to user-specified muscle activation patterns. A series of meshes was created to perform a mesh validation study; in the validation tests performed, a 245,223-element mesh was found to be sufficient to model tongue behavior.

Systematic samples of the behavior of the model were collected. Principal component analyses were performed on the samples to discover low-dimensional representations of tongue postures. Statistical models (linear regression models and neural networks) were fit to predict tongue posture from muscle activation, and vice versa. In all tests, it was found that a relatively small sample of tongue postures can be used to successfully generalize to larger data sets.

Finally, a variety of specific tests were performed, based on claims and predictions found in previous literature. Of these, the claims of the muscular hydrostat theory of tongue movement were best supported. Simulations were also run that simulated lingual hemiplegia. It was found that substantially different muscular activation patterns were required to achieve equivalent postures in a hemiplegic tongue, relative to a normal tongue.

The present work advances the state of tongue modeling in several respects. It is based on a detailed anatomical study of an actual specimen. It is also the first tongue model to have been subjected to a mesh validation study. It is anticipated that the model will be of use in further research into the mechanical properties of the tongue, and the relationship between tongue muscle activation and the acoustical output of the vocal tract.

All contents copyright © 2008 Adam Baker