摘要
该文研究了在基于矢量量化的说话人识别方法中采用加权的失真测度对识别率的影响。在采用加权欧氏距离失真测度时,利用特征参数的离散程度来确定权值,提出了基于标准差的加权失真测度和基于方差的加权失真测度。实验结果表明,在以MFCC为特征参数的说话人识别系统中,使用这两种算法均可以提高识别率。
This article researches the effect of weighted distortion measure to recognition rate in the way of VQ speaker recognition. It uses dispersion degree of the characteristic parameter to ensure the value of weight, and advances weighted distortion measure based on standard deviation and variance. The experiment shows that the two method can increase the correct rate of speaker recognition using 'MFCC' as characteristic parameter.
作者
李娜
崔连延
李兰兰
LI Na, CUI Lian-yan, LI Lan-lan (Liaoning University of Technology, School of Electronics and Informatics Engineering, Jinzhou 121000, China)
出处
《电脑知识与技术》
2008年第12Z期2228-2229,共2页
Computer Knowledge and Technology
关键词
说话人识别
矢量量化
标准差
方差
speaker recognition
vector quantization
standard deviation
variance
