Abstract:

A physiological articulatory model was developed for simulation of 3-D deformation and movement of speech organs. The model is composed of a midsagittal portion of the tongue and the jaw and hyoid bone that yield rotation and translation. The tongue model consists of a midsagittal layer and two parasagittal layers with 1-cm intervals. The form and muscle geometry of the whole model are extracted from MR images of a speaker. All the soft tissue and rigid organs are modeled by node points with mass and two types of branches to adjacent points: soft branches for muscles and connective tissue between node points, and hard branches for bony organs. The muscles are defined based on an improved Hill's model. Muscle activation signals, in part derived from EMG recordings, serve as the model's control parameters. Preliminary experiments revealed that the model can produce realistic 3-D deformation and movement for vowels and a sustained consonant /r/. Since all the displacements of soft tissue and rigid organs are computed with the same algorithm, the model produces plausible dynamic patterns of the tongue--jaw complex with relatively short computational time in comparison with the finite-element method.