Co-articulation Generation Using Maximum Direction Change and Apparent Motion for Chinese Visual Speech Synthesis

Abstract

This study presents an approach for automated lip synchronization and smoothing for Chinese visual speech synthesis. A facial animation system with synchronization algorithm is also developed to visualize an existent Text-To- Speech system. Motion parameters for each viseme are first constructed from video footage of a human speaker. To synchronize the parameter set sequence and speech signal, a maximum direction change algorithm is also proposed to select significant parameter set sequences according to the speech duration. Moreover, to improve the smoothness of coarticulation part under a high speaking rate, four phonemedependent co-articulation functions are generated by integrating the Bernstein-Bézier curve and apparent motion property. A Chinese visual speech synthesis system is built to evaluate the proposed approach. The synthesis result of the proposed system is compared to the real speaker. The coarticulation generated by the proposed approach is also evaluated.