基于GMM的说话人识别系统研究及其MATLAB实现  

Research on Speaker Recognition System Based on GMM and Its Implementation in MATLAB

作　　者：何建军 HE Jian-jun(Technology Center,Shenzhen Electrical Appliances Company,Shenzhen 518001,China)

机构地区：[1]深圳电器公司技术中心,广东深圳518001

出　　处：《软件导刊》2021年第8期49-57,共9页Software Guide

摘　　要：为在嵌入式平台上实现说话人识别,分析研究说话人语音信号预处理、特征提取及GMM模型基本原理,并应用MATLAB实现基于GMM模型的说话人识别系统。基于TIMIT语料库,通过调整GMM阶数和语音时长,对系统性能进行验证分析。实验结果表明:①随着GMM模型阶数的增加,识别率随之提升,但计算量也急剧增加,当阶数达到16附近时,识别率则不再提升,反而出现了降低的趋势;②增加训练样本时长可从总体上提升识别率,但达到一定程度后便很难再继续提升。该结果对于在嵌入式平台上实现说话人识别具有较高参考价值。In order to realize speaker recognition on the embedded platform,analyzes and studies the basic principles of speech signal preprocessing,feature extraction and GMM model,and uses MATLAB to design a speaker recognition system based on the GMM model.Then based on the TIMIT corpus,the performance of the speaker recognition system based on the GMM model is verified and analyzed by adjusting the order and the speech duration of the GMM model.The experimental results show that:①As the order of the GMM model increases,the recognition rate also increases,but the computational workload also increases sharply.When the order reaches around 16,the recognition rate will not increase,but decrease;②The recognition rate can be increased by increasing the training sample duration as a whole,but it is difficult to increase again after a certain duration.The experimental result has certain reference significance for the realization of speaker recognition on the embedded platform.

关 键 词：说话人识别 语音识别 美尔频率倒谱系数 高斯混合模型 MATLAB 

分 类 号：TP303[自动化与计算机技术—计算机系统结构]