基于分类器与阵列搜索算法的文本无关说话人识别  

Text-Independent Speaker Recognition Based on Classifier and Array Searching

作　　者：戚银城[1] 柴君[1] 赵振东[1] 

机构地区：[1]华北电力大学电子与通信工程系,河北保定071003

出　　处：《微电子学与计算机》2009年第4期43-46,共4页Microelectronics & Computer

摘　　要：基于小波神经网络的原理对与文本无关的说话人进行了识别.通过分析网络结构,提出了优化网络初始参数的方法,加快了小波网络的收敛速度;采用分组器与阵列搜索算法相结合,减少了识别时的搜索时间.实验结果表明该识别系统能大大提高识别人数和识别速度,在43人的语音识别中,识别率达到97.67%.The principle of wavelet neural network is applied to text-independent speaker recognition. By analysising of the network structure, we proposed a method to optimize the initial values of wavelet neural networks, which has a faster con- vergence speed. By using a preclassification and array searching algorithm, searching time is saved. Experimental results show that the system has a good performance in speaker numbers and recognition speed, and the recognition rate reaches to 97.67% for 43 speakers.

关 键 词：说话人识别 小波神经网络 初始值 阵列神经网络 

分 类 号：TN912[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]