孤立词识别系统的算法改进及优化  

Improvement and Optimization of Algorithm of Isolated Word Recognition System

作　　者：朱健晨 刘增力[1] 袁洪[2] 程桐 

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500 [2]昆明冶金高等专科学校,云南昆明650093 [3]哈尔滨工程大学国家保密学院,黑龙江哈尔滨150001

出　　处：《计算机仿真》2015年第9期249-253,310,共6页Computer Simulation

基　　金：国家基金项目(61271007)

摘　　要：针对基本孤立词识别系统中语音信号预处理效果差、模板匹配成功率低、词汇识别耗时长等问题,基于动态时间规整算法(Dynamic Time Warping,DTW)提出了一种改进孤立词识别系统。首先,通过仿真设置合理的帧长和帧移数,并利用改进的端点检测法确定语音的始末端,提高了语音信号的预处理效果;其次,采用美尔倒谱系数结合一阶差分系数提取了语音信号的特征参数,从而有效降低了计算机的时间复杂度;然后,采用DTW算法有效降低了语音信号的累积失真距离,实现了对孤立词语音信号的有效识别;最后,通过计算机仿真,验证了所提出的设计及参数设置的有效性,并通过对比实验验证了设计的高效性。In this paper, we put forward an improved isolated word recognition system based on the dynamic time warping algorithm （Dynamic Time Warping, DTW）. First, we set reasonable frame length and shift through the sim- ulation experiment, and determined the beginning and end terminal of the speech by Endpoint detection method, which can improve the pretreatment effect of speech signal. Second, we extracted characteristic parameters of the speech signals by the method of Mel Frequency Ceptral Coefficients combined with the first order differential coeffi- cient, which can reduce the time complexity of the computer effectively. Then, we realized effective recognition of i- solated word speech signal by DTW algorithm, which can shorten the distorted distance accumulated by the speech signal effectively. Finally, we demonstrated the effectiveness of the method and parameter settings mentioned above through the simulation experiment and verify high-efficiency of the method through contrast experiment.

关 键 词：模板匹配 端点检测算法 特征参数提取 识别率 识别速度 

分 类 号：TN802[电子电信—信息与通信工程]