语音识别中DTW特征匹配的改进算法研究  

A Research on Improving the Algorithm for DTW Feature-Matching in Speech Recognition

作　　者：刘长明[1] 任一峰[1] 

机构地区：[1]中北大学自动控制系,山西太原030051

出　　处：《中北大学学报（自然科学版）》2006年第1期37-40,共4页Journal of North University of China(Natural Science Edition)

摘　　要：采用基于动态规划方法的动态时间归正技术DTW(D YNAM IC T IM E W ARP ING),可成功解决语音信号特征参数序列比较时时长不等的问题.在基于DTW的特征匹配用改进的动态时间归正方法将模板特征序列和语音特征序列进行匹配的基础上,比较两者之间的失真,得出识别判决的依据.实验表明,改进后的算法在孤立词语音识别中获得了良好性能.The dynamic time warping （DTW） technology based on the dynamic programming can solve the problem that the characteristic parameters of speech signal are of different lengths in each other. Matching the pattern character and speech character with dynamic time warping enables the researchers to compare the distortions between them so as to obtain the criterion for identification. According to the experimental results, the algorithm after improvement performs better in single speech recognition.

关 键 词：动态规划(DP) 动态时间归正(DTW) 模板匹配 

分 类 号：TN912.3[电子电信—通信与信息系统]