基于小波变换的说话人识别技术  

Research on Speaker Recognition Based on Wavelet Transform

作　　者：檀蕊莲[1,2] 柏鹏[2] 李哲[2] 姚战宏[2] 栾前进 

机构地区：[1]武警工程大学信息工程系,陕西西安710086 [2]空军工程大学装备发展与运用研究中心,陕西西安710051 [3]总参陆航部驻西安地区军代室,陕西西安710000

出　　处：《空军工程大学学报（自然科学版）》2013年第1期85-89,共5页Journal of Air Force Engineering University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61174194)

摘　　要：说话人识别技术是通过判断待识别人语音与预先提取的说话人语音特征是否匹配来鉴别说话人身份的一种生物认证技术,环境噪声是说话人识别技术走向实用化的一个主要障碍。针对噪声环境中说话人识别性能较差的不足,结合小波变换的优点,提出了将小波变换技术与传统的特征参数提取方式相结合的方法。该方法首先对语音信号进行小波分解,在此基础上再对小波系数进行阈值处理,仅保留阈值以上的数据,而后提取相关性不大的传统特征参数进行组合,分别作为说话人识别系统的输入矢量。仿真结果表明:在噪声环境中,说话人识别系统能较好识别出说话人,经过小波变换后再提取特征参数的方法可以得到更高的识别率,大大提高说话人识别系统的识别性能。Speaker recognition is a kind of biological authentication technology which distinguishes speakers＇ identity by matching the voice distilled beforehand.However,the noise circumstance is an obstacle disturbing this technology walking up to practicality.Concerning the shortcoming of poor speaker recognition performance in noisy environments and combining the advantages of wavelet transform,a method of combining the wavelet transform technology with the traditional characteristic parameter extraction mode is proposed.In this method,the speech signal is decomposed by the wavelet,and then wavelet coefficients are processed by threshold.Only the data above the threshold are retained.The traditional characteristic parameters of little correlation are extracted to use as the input vector of the speaker recognition system.The simulation results indicate that the use of the method can better identify the speaker.A higher recognition rate can be obtained through the wavelet transform first and then the extraction of characteristic parameters.The application of this method greatly improves the performance of the speaker recognition system

关 键 词：说话人识别 动态时间规整 矢量量化 小波变换 组合特征 

分 类 号：TN912.3[电子电信—通信与信息系统]