基于DTW改进算法的孤立词语音识别仿真  

Design of isolated words speech recognition system based on improved DTW algorithm

作　　者：刘静[1] 王儒[1] 曲金玉[1] 李东荣[1] 

机构地区：[1]山东理工大学交通与车辆工程学院,山东淄博255091

出　　处：《山东理工大学学报（自然科学版）》2013年第1期63-66,共4页Journal of Shandong University of Technology:Natural Science Edition

摘　　要：为提高识别率和识别效率,采用双门限过零率和短时能量作为端点检测的依据,提取Mel频率倒谱系数作为语音特征参数,并使用DTW算法进行模式匹配.由于传统DTW算法计算量大,所以采用局部路径约束和区域约束进行改进,并用Matlab对改进后的DTW算法进行了仿真.实验证明该算法对孤立词语音识别能够达到较好的识别结果.To improve recognizing rate and recognizing efficiency, double-threshold zero-crossing rate is adopted in the endpoint detection, Mel-Frequency Cepstral Coefficients is obtained as speech characteristic parameters, and DTW algorithm is used for matching of the model. For the large amount of data in the matching process, DTW algorithm is improved by using local con-straints and global constraints and it was simulated by Matlab. The experiment shows that this algorithm can achieved good results in isolated word speech recognition .

关 键 词：孤立词 语音识别 MATLAB DTW 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]