Implementation Of Tibetan Language TTVS Based On Xface

TTVS(Text-to-Visual Speech) technology has become a research hot spot with the rapid development of human-computer speech interaction. However, the TTVS for Chinese minorities are still in its infancy. Therefore the research on Tibetan TTVS has groundbreaking. The thesis proposes a method to realize the Tibetan TTVS based on the Xface platform. A visual 3D facial animation system is established according to the MPEG-4(Moving Pictures Experts Group-4) standard. The facial action is driven by a Tibetan mouth shape library that is obtained by analyzing the characteristics of Tibetan initials and Tibetan finals. The 3D face image is combined with a Tibetan statistical parametric speech synthesis system to realize a Tibetan TTVS. The main works and originalities of the thesis are as follows:Firstly, Xface-based facial animation system is realized by using a face model that uses facial definition parameter(FDP) and facial animation parameter(FAP) in MPEG-4 standard to describe the face image and face animation. A MPEG-4 based facial system is set up by using XFace platform.Secondly, a Tibetan TTVS oriented Tibetan face model is established in the thesis. The initials and the finals of Tibetan are used as unit to define and describe the mouth shape animation of Tibetan pronunciation. The description and definition of mouth shape for FDP and FAP of Tibetan pronunciation are realized according to the similarity of initial and final pronunciation between Mandarin and Tibetan. The mouth shapes of pronunciation for those initials and finals which are same with Mandarin are defined by using the mouth shape definition of the International Phonetic Alphabet(IPA). The mouth shapes of pronunciation for those initials and finals which are different from Mandarin are defined according to their characteristic of pronunciation. The Tibetan face model is established by using Xface with FDP and FAP of mouth shape of Tibetan initials and Tibetan finals.Finally, a Tibetan TTVS is realized based on Xface platform. The time boundaries of each initials and finals in facial animation system are aligned with the boundaries of each initials and finals in Tibetan TTS so that the pronunciation can match the mouth shape. The Tibetan TTVS is realized by combining the Tibetan facial animation and Tibetan TTS. The system is evaluated in terms of visualization effect and synthesized speech quality with and without facial animation. Tests show that the mean MOS for visualization effect is 3.6. The mean MOS for synthesized Tibetan speech without facial animation is 4.1 while the mean MOS for synthesized Tibetan speech with facial animation is 4.6.