基于VQ的与文本无关的说话人确认系统  

Text-independence speaker identification system based on vector quantization

作　　者：陈明义[1] 周昆湘[1] 曾理文[1] 

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《信息技术》2007年第3期97-98,127,共3页Information Technology

摘　　要：通过介绍说话人确认系统,采用能够反映人对语音的感知特性的Mel频率倒谱系数(MFCC)作为特征参数,同时提出了用训练得到的码本以及平均失真作为表征说话人的模型,对说话人确认系统中距离测度做出了相应的改动。实验结果表明系统具有较好的性能。This paper introduces a speaker identification system, which uses weighted Mel - frequency cepstral coefficients as the feature parameters, and proposes a new model which uniting VQ coder and the quantization distortion to represent a particular speaker. Based on this model, the method of calculating the distance of speaker identification system based on VQ has been changed. The experiment shows that it has better performance.

关 键 词：说话人确认 矢量量化 MFCC 文本无关 

分 类 号：TN912.3[电子电信—通信与信息系统]