一个语音驱动的人脸合成系统  

A FACE SYNTHESIS SYSTEM DRIVEN BY VOICE

作　　者：孙岭[1] 戴礼荣[1] 赖伟[1] 王仁华[1] 

机构地区：[1]中国科学技术大学电子工程与信息科学系,合肥230027

出　　处：《模式识别与人工智能》2003年第4期459-464,共6页Pattern Recognition and Artificial Intelligence

摘　　要：本文采用多层前馈神经网络,对汉语中的声母和韵母的发音唇型参数及其相应的语音参数之间的映射进行了研究.通过实验,对于在进行语音参数和唇型参数的映射研究中,选择哪种语音参数更为合适进行了分析.最后,把该网络应用于一个人脸合成系统。该系统能够实时地合成和语音同步而且较为自然的唇型.In recent years, visual speech has received a lot of attention and played an active role in computer-human interactive technology. A large effort has been directed to mapping speech to lip movement and much attention has been paid to lip synchronization in face synthesis research. In this paper, a multi-layer feed-forward neural networks is used to map the speech of initials and finals in mandarin to corresponding lip movement. By experiments, which kind of speech parameters is more suitable for the mapping is analysed. The trained network is then applied to a visual speech system. The system can synthesize natural lip movement synchronized with audio speech.

关 键 词：人脸合成系统 语音驱动 前馈神经网络 唇型参数 语音参数 隐马尔可夫模型 人脸建模 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]