A Novel Method for Constructing 3D Geometric Articulatory Models

Abstract

This study describes a novel method of constructing a geometric articulatory model based on magnetic resonance imaging data by taking the physiological boundaries of speech apparatus into account. Two improvements have been made to the modeling process: i) Images taken from different viewpoints are combined to improve the accuracy of outline annotation. ii) Speech organs' meshes are modeled with reference to the anatomical structures. Both qualitative and quantitative evaluations indicated that the proposed method surpasses the conventional method. Based on the meshes of the speech organs associated with different articulations, the linear component analysis was used to extract the control parameters. Each speech organ can be described using three control parameters or fewer. After the reconstruction, the average error between model and real data was less than 1.0 mm. This is also the first effort made to construct a 3D vocal tract model based on Chinese MRI data. It will facilitate the theoretical study and practical use in Chinese-speech-production related issues.