摘要
人脸语音动画是虚拟现实领域的热点,语音特征参数提取是实现语音同步动画的前提和关键所在。为了能够提取鲁棒性更好的语音特征参数,在小波变换的理论基础上,借鉴MFCC特征参数的提取方法,运用表征语音动态特征的特征差分算法,提出了一种基于离散小波变换的语音特征参数(DWTMFCC)提取方法,并与反映语音情感特征的韵律参数相结合。通过基于LGB算法的VQ模型进行说话人语音识别,可以得到组合特征参数的识别率较高。
Facial speech animation is a popular issue in the area of virtual reality. The speech feature extraction is the premise and key to realize the speech synchronous animation. In order to extract the pronunciation feature parameters whose robustness is better, on the basis of wavelet transform theory and the extraction method of MFCC reference of characteristic parameters, a phonetic feature parameter extraction method based on discrete wavelet transform is proposed by the aid of a feature difference algorithm which characters the speech dynamic features. It combines the prosodic parameter which reflects the characteristics of speech emotion and acquires high recognition rate of mixed characteristic parameter. The speaker＇s voice recognition is implemented with the VQ model based on LGB algorithm.
出处
《现代电子技术》
2011年第6期74-77,共4页
Modern Electronics Technique
基金
国家"863"高技术研究发展计划(2007AA01Z324)
关键词
人脸语音动画
语音特征提取
小波变换
动态特征
组合特征参数
facial speech animation
extraction of speech feature
wavelet transform
dynamic feature
mixed characteristic parameter
