14th International Congress of Phonetic Sciences (ICPhS-14)

San Francisco, CA, USA
August 1-7, 1999

A Computational Model for 2D Articulation: Speech Production with Potential Use in Recognition

Leo J. Lee (1), Jianwu Dang (1,2), Li Deng (1)

(1) Department of Electrical & Computer Engineering, University of Waterloo, Waterloo, ON, Canada
(2) ATR Human Information Processing Research Labs, Kyoto, Japan

This paper describes a 2D physiological model of the vocal tract to simulate the dynamic actions of speech production organs including the tongue, mandible, hyoid bone, and the outer wall of the vocal tract. This model has drastically simplified a previous 3D physiological model into an efficient computational form. The movement of the articulators is driven by the phonedependent, slowly time-varying targets specified in the articulatory space. Appropriate muscle activation signals are automatically generated by comparing the current and target position of the articulators. The smooth trajectories of the articulators thus produced are converted to the time-varying vocal tract area function as the final output of the model. Finally, we discuss the potential use of this simplistic but efficient 2D production model in the state-space or segmental modeling approach to automatic speech recognition, where the articulatory model serves as an underlying hidden speech production state equation with realistic physiological constraints.

Full Paper

Bibliographic reference.  Lee, Leo J. / Dang, Jianwu / Deng, Li (1999): "A computational model for 2d articulation: speech production with potential use in recognition", In ICPhS-14, 2529-2532.