虚拟人“双簧”—与语音同步的三维人脸动画的研究

COLLABORATE TWO-VIRTUAL HUMAN SHOW—ON 3D FACIAL ANIMATION SYNCHRONISED WITH SPEECH

为了有效地合成人脸语音动画,提出一种与语音同步的三维人脸口型与表情动画合成的方法。首先,根据对人脸运动解剖学的研究,构建一个基于肌肉模型与微分几何学模型的三维人脸控制模型,通过数据结构的形式去控制肌肉模型和微分几何学模型实现人脸的运动,从而实现各种口型和表情运动的变化;然后,充分考虑汉语的发音特征,提出一种基于几何描述的并且符合汉语发音习惯的协同发音模型,从而产生逼真的三维人脸语音动画。仿真结果和实验对比表明,采用该方法可以得到符合汉语发音习惯的三维人脸口型动画,且合成的三维人脸表情较为自然,逼真。

In order to effectively synthesise the 3D face and speech animation,we proposed a method of animation synthesis with speech synchronisation for 3D mouth shapes and facial expressions. First,according to the research on anatomy of facial movement,we built a 3D face control model,which is based on muscle model and differential geometry model,it controls the muscle model and differential geometry model in the form of data structure to realise facial movement,so that implements the variation of mouth shapes and expression movements.Then,fully considering the characteristics of Chinese pronunciation,we presented a coarticulation model,which is based on geometric description and compliant with Chinese pronunciation habit,so that the vivid 3D face animation with speech could be generated. Finally,it was indicated by the comparison between simulation result and experiment that to apply this method could realise the 3D facial mouth shapes animation in accordance with the habits of Chinese pronunciation,and the synthesised 3D facial expressions were more natural and vivid.