基于元音MFCC的说话人识别系统研究  

Study of Speaker Recognition System Based on Vowel MFCC

作　　者：应武[1] 

机构地区：[1]金华职业技术学院,讲师浙江金华321017

出　　处：《电子测量与仪器学报》2007年第3期48-51,共4页Journal of Electronic Measurement and Instrumentation

摘　　要：说话人识别从本质上看是从语音信息中提取说话人特征,并通过一定的方式进行模式识别的过程。辨别说话人的方法很多,本文认为先从语音中提出元音,再通过计算元音的MFCC(美尔频标倒谱系数)特征参数,并与DTW(动态时间规整)结合进行多人多单词试验,实验证明这种识别方式能提高识别率5%左右——从原字平均识别率为83%提高到取元音后平均识别率为88%。In essence, speaker recognition is a process of extracting speaker＇s features from speech information and carrying out pattern recognition through certain method. There are many ways to identify the speaker. This paper proposes that we first extract vowels from the speech sounds, calculate MFCC （Mel Frequency Cepstral Coefficient） parameters of the vowels, which are combined with DTW （Dynamic Time Warping）, and then carry out multi person, multi word trial. Experiment result shows that the proposed recognition method can improve the recognition rate about 5%, that is, after extracting vowels the average word recognition rate increases from 83% to 88%.

关 键 词：说话人识别 元音提取 MFCC特征参数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]