语音识别中基于SFCM模糊聚类的矢量量化方法  

A VECTOR QUANTIZATION APPROACH BASED ON SFCM FUZZY CLUSTERING IN SPEECH RECOGNITION

作　　者：李晶皎[1] 孙杰[1] 姚天顺[1] 

机构地区：[1]东北大学信息科学与工程学院

出　　处：《计算机研究与发展》1999年第3期263-267,共5页Journal of Computer Research and Development

基　　金：国家自然科学基金;国家教委博士点基金

摘　　要：模糊聚类分析算法用隶属度确定样本所属类别，因其良好的效果而被广泛用于语音识别领域．文中提出了一种模糊聚类分析算法ＳＦＣＭ，并将其用于语音特征的矢量量化，最终形成码本尺寸为１２８的码本．用ＳＦＣＭ算法得到的码本分布合理，没有空类．采用此码本的语音识别实验表明了这种量化方法对语音识别的有效性．The algorithm of fuzzy clustering analysis can be used to determine the sample classification.Because of its good effect,this method has been adopted widely in the field of speech recognition.Here presented is a fuzzy clustering analysis algorithm SFCM,which is applied to the vector quantization of the speech feature.The code vector that is 128 quantization degrees is built.The distribution of the code vector that is obtained by SFCM is reasonable and there are not empty classes.The experiment results of the speech recognition that employs this kind of code vector demonstrate the efficiency of this quantization method for speech recognition.

关 键 词：模糊聚类分析 语音识别 矢量量化 SFCM 

分 类 号：TN912.34[电子电信—通信与信息系统]