在孤立词语音识别中动态时间规正的改进算法  

Improved DTW algorithm on isolated-word speech recognition

作　　者：汲清波[1,2] 卢侃[2] 李康 

机构地区：[1]北京航空航天大学自动化科学与电气工程学院,北京100191 [2]哈尔滨工程大学信息与通信工程学院,哈尔滨150001

出　　处：《计算机工程与应用》2010年第25期118-120,共3页Computer Engineering and Applications

基　　金：中央高校基本科研业务费专项资金No.HEUCF100826~~

摘　　要：动态时间规正(Dynamic Time Warping)是语音识别中的一种经典算法,算法简单有效,因此在实现孤立词识别系统中获得了广泛的应用。提出一种DTW的改进算法,采用两次在时间域上的规正,使计算程序简洁规范,计算量减少。经实验验证,改进DTW算法在不降低识别率的前提下,提高了识别速度,提高了系统的实用性。Dynamic Time Warping is a kind of classical algorithm in speech recognition system.Because the algorithm is simple and valid, it achieves extensive application in speech recognition of single word.This paper raises improved DTW algorithm with twice time warping, so that it wiU make program normative and reduce the calculation work.Analysis of the simulation results shows that the improved DTW algorithm not only can increase recognition speed but also improve practicability of speech recognition system,without reducing recognition rate.

关 键 词：语音识别 动态时间规正算法 相似度 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]