用于人脸动画的语音特征提取算法研究  

Research for Speech Feature Extraction Algorithm During Visual Speech Synthesize

作　　者：刘颖[1] 王成儒[1] 

机构地区：[1]燕山大学信息科学与工程学院电子与通信工程系,河北秦皇岛066004

出　　处：《电声技术》2008年第12期49-53,共5页Audio Engineering

摘　　要：针对人脸语音动画提出了基于小波包分析的语音特征提取算法,为了表征语音的动态特性,采用了特征差分和基于口形帧前后关联的多帧语音特征提取,并利用主成分分析(Principle Component Analysis,PCA)降低了输入语音的特征维数。基于隐马尔可夫模型(Hidden Markov Model,HMM)的可视语音合成系统的实验表明该算法提取的语音参数比传统的Mel频率倒谱系数(Mel-Frequency Cepstrum Coefficients,MFCC)合成的口形效果要好,有利于可视语音合成的研究。Speech feature extraction algorithm based on wavelet packet analysis is proposed for visual speech synthesize. In order to express dynamic characteristic for speech, feature difference and multi-frames speech feature extraction based on association relationship of lip frames are introduced. PCA（Principle Component Analysis） is utilized to reduce dimension of the input speech. Visual speech synthesize system based on HMM（Hidden Markov Model） indicates that the effect of lip synthesize using the proposed algorithm is better than traditional MFCC （Mel-Frequency Cepstrum Coefficients） and it is useful for the research of visual speech synthesize.

关 键 词：语音特征提取 小波包分析 人脸语音动画 主成分分析 

分 类 号：TN912[电子电信—通信与信息系统]