首页 | 本学科首页   官方微博 | 高级检索  
     检索      

一种基于Fisher准则的说话人识别方法研究
引用本文:张芸,李昕,郑宇,杨庆涛.一种基于Fisher准则的说话人识别方法研究[J].兰州大学学报(自然科学版),2007,43(2):92-97.
作者姓名:张芸  李昕  郑宇  杨庆涛
作者单位:1. 上海大学,机电工程与自动化学院,上海,200072
2. 上海大学,机电工程与自动化学院,上海,200072;南京大学,电子科学与工程系,江苏,南京,210093
3. 上海大学,计算机学院,上海,200072
基金项目:上海市教委资助项目,上海市科委资助项目
摘    要:在说话人识别系统中,提高反映说话人个性的语音信号特征参数的有效性和实时性是问题之一.本文在使用线性预测系数倒谱(LPCC)和美尔倒谱系数(MFCC)计算特征参数的基础上利用Fisher准则,构造了一种新的混合特征参数.这种新的参数在不增加系统计算量的同时,结合了LPCC和MFCC各自的优点,具有更好地表征说话人特征的能力,并在一定程度上消除特征的信息冗余,有利于信息的实时处理.

关 键 词:说话人识别  Fisher准则  特征参数降维
文章编号:0455-2059(2007)02-0092-06
修稿时间:2005-12-30

Speaker recognition based on Fisher discriminant criterion
ZHANG Yun,LI Xin,ZHENG Yu,YANG Qing-tao.Speaker recognition based on Fisher discriminant criterion[J].Journal of Lanzhou University(Natural Science),2007,43(2):92-97.
Authors:ZHANG Yun  LI Xin  ZHENG Yu  YANG Qing-tao
Institution:School of Electromechanical Engineering and Automation, Shanghai University, Shanghai 200072, China; 2. Department of Electronics Sciences ~ Engineering, Nanjing University, Nanjing 210093, China; 3. School of Computer Science and Technology, Shanghai University, Shanghai 200072, China
Abstract:How to extract the valid acoustic features which has good representation and real-time probability is one of the key problems in speaker recognition. A novel feature extraction method for speaker identification is proposed in this paper. The feature vectors are constructed through combining linear prediction cepstrum coefficients and Mel-frequency cepstrum coefficients. Then the dimensional down was completed using the Fisher discriminant criterion. The experiment results show that the mixed feature vectors with dimensional down efficiently improved the speaker identification accuracy with better feature characterization.
Keywords:speaker recognition  Fisher discriminant criterion  feature vectors dimensional down
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号