Minimizing Vertices in 3D Lips Model Using Relevance Measure

Abstract

This paper presents works involved in developing a three-dimensional lips in a visual speech synthesis system. Several works have been done in similar areas, but a systematic approach of identifying and determining appropriate number of vertices in the lips model were not explained thoroughly. Instead of the development technique, this paper is also presents an experimental set up designed to test the capabilities and flexibility of the lips in deforming speech. This paper explains on the process of capturing the lips and on the relevance measure technique used to identify optimum number of vertices to deform visemes in Standard Malay language. Results shown in this work proves that a systematic degradation by retaining 0.005% of 24,000 vertices into 120 vertices would remain the similar quality of lips deformation. Therefore, the outcome of this study justifies that an appropriate number of vertices which suitable to be used in visual speech synthesis for 3D lips model.