摘要： 利用HMM的统计特性,对HMM模型结构进行改动,使其成为人脸语音动画合成中语音特征到图像特征的映射模型。通过一些必要的前期处理,就可以根据训练样本建立特定说话对象的HMM。使用该模型,加上一些必要的后期处理工作,就可以通过输入的语音信号合成语种无关的、平滑的、并富有真实感的人脸语音动画。

关键词: 语音动画, 隐马尔可夫模型, 预处理

Abstract: This paper takes advantage of HMM’s statistic characters. With a little modification on HMM structure, it gets a mapping model from speech to image. After some pre-process, a HMM of certain object can be created according to training samples. Using this model , together with some necessary post-process work, it can synthesize smooth, realistic face speech animation independent of classification of language only with speech signal.

中图分类号: 

• TN912.34

• TP391.41