一种基于DTW的孤立词语音识别算法  

A Method of Isolated Word Recognition Based on DTW

作　　者：张军[1] 李学斌[1] 

机构地区：[1]北京化工大学信息科学与技术学院,北京100029

出　　处：《计算机仿真》2009年第10期348-351,364,共5页Computer Simulation

摘　　要：针对动态时间规整(DTW)对孤立词端点检测准确性过度依赖的问题,针对上述问题,采用放宽端点和限定动态规整计算范围结合的算法,不仅更准确的放松前后端点降低端点检测的敏感度,而且结合对动态规整计算范围的限定,减少计算量,提高执行效率。分别测试了基于传统DTW算法的识别率和改进后DTW算法的识别率。实验结果表明,改进后的算法,能有效提高孤立词识别率。The resuh of isolated word recognition of the DTW depends much on the accuracy of endpoint detection. In order to reduce the influence, a modified method of endpoint detection, for relaxing the endpoint of speech signals combining with the dynamic rule computing range restriction is proposed to overcome this disadvantage. It can not only more accurately relax the forward and back endpoint to reduce its sensitivity, but also reduce the amount of calculation and improve the performance with the dynamic rule computing range restriction. Compared the traditional DTW with modified DTW algorithm, the results show that the modified algorithm can provide a better performance in the speech recognition rate.

关 键 词：语音识别 动态时间规整 端点检测 孤立词 

分 类 号：TN912.3[电子电信—通信与信息系统]