基于语音情感识别的多表情人脸动画方法  

Multi-expression Facial Animation Based on Speech Emotion Recognition

作　　者：谢金晶[1] 陈益强[1] 刘军发[1] 

机构地区：[1]中国科学院计算技术研究所普适计算研究中心,北京100190

出　　处：《计算机辅助设计与图形学学报》2008年第4期520-525,共6页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(60775027;60575032);北京市科技新星计划基金(2005B54)

摘　　要：基于MPEG-4标准,实现了一种由彩铃语音及蕴含情感共同驱动生成人脸动画的方法和系统.选用HMM作为分类器,训练使其识别语音库中嗔怒、欣喜、可爱、无奈和兴奋5类情感,并对每类情感建立一组与之对应的表情人脸动画参数(FAP).分析语音强弱得到综合表情函数,并用此函数融合表情FAP与唇动FAP,实现人脸表情多源信息合成,得到综合FAP驱动人脸网格生成动画.实验结果表明,彩铃语音情感识别率可达94.44%,该系统生成的人脸动画也具有较高的真实感.A facial animation generation system is implemented based on MPEG-4 standard according to both multimedia ring and the emotion embedded. By choosing hidden Markov model as the classifier, five kinds of emotions, including anger, gratulation, cute, puzzle, excitement, are generated in speech database through training, and then a set of expression FAPs for each kind of emotion is built up. By analyzing the intensity of speech signal, synthesis expression function is obtained and used to integrate expression FAP with lip animation FAP to obtain the synthesis FAP for driving facial mesh to generate animation. The test result indicates that the maximum recognition ratio of our system may achieve 94.44 %, besides its realistic facial animation.

关 键 词：MPEG-4标准 语音情感识别 隐马尔可夫模型 人脸动画 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]