一种基于差别子空间的语音识别算法研究和实现  

The study and implementation of a speech recognition algorithm based on difference subspace

作　　者：张歆奕[1] 吴今培[1] 张其善[2] 

机构地区：[1]五邑大学电子与信息工程系,广东江门529020 [2]北京航空航天大学电子工程系,北京100083

出　　处：《五邑大学学报（自然科学版）》2002年第1期17-20,共4页Journal of Wuyi University(Natural Science Edition)

摘　　要：介绍了一种基于差别子空间的语音识别算法,并从最优标准和最优解的角度,在理论上论证了该算法与传统的DTW算法的优劣. 然后用MATLAB实现了这两种算法,并进行了大量的孤立词语音识别实验. 理论和实验表明,基于差别子空间的语音识别算法非常有效,识别率在多次训练时高于传统的DTW算法.A speech recognition algorithm based on difference subspace is in this paper first introduced, then a comparison is made theoretically between this algorithm and the traditional DTW algorithm from the view point of optimization criteria and optimization solution. The two algorithms are implemented using MATLAB, and experiments of isolated-word recognition are done. Both theoretical analysis and experimental results show that the algorithm based on difference subspace is more efficient, and that its recognition rate is higher than DTW when training times are high.

关 键 词：差别子空间 语音识别 DTW算法 MATLAB 模板匹配 共性矢量 剩余矢量 

分 类 号：TN912.34[电子电信—通信与信息系统]