基于倒谱加权的说话人识别技术  

Speaker recognition technology based on cepstral coefficients weight

作　　者：孙林慧[1] 李平安[1] 

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《西安邮电学院学报》2006年第3期9-12,共4页Journal of Xi'an Institute of Posts and Telecommunications

基　　金：江苏省高校"青蓝工程"计划项目(QL003YZ)

摘　　要：提出一种通过加不同权值来考察语音各阶参数的抗噪性强弱的方法,采用此方法分析了对干净语音和加白噪声语音的Mel倒谱各阶参数加不同权值后对识别系统性能的影响。得出结论:高阶Mel倒谱参数抗噪性较差而低阶倒谱参数抗噪性较好。因此可以根据抗噪性的强弱对该倒谱各阶参数进行合理加权,以便突出抗噪性强的参数对说话人识别系统的贡献。A method for researching the noise- resistance characters of each order of Mel eepstral coefficients by choosing different weights is presented. Based on it, the noise - resistance characters of each order of Mel cepstral coefficients are analyzed. The results show that lower order has better robustness than higher order, so the emphases can be put on the contribution of lower order coefficients by choosing appropriate weight.

关 键 词：权值 抗噪性 Mel倒谱 说话人识别 

分 类 号：TN912.3[电子电信—通信与信息系统]