一种改进的聚类算法及其在说话人识别上的应用

目前应用最广泛的模糊聚类算法是基于目标函数的模糊k-均值算法．针对该算法存在的缺点。本文提出一种改进的聚类算法．利用遗传算法的全局优化的特点，在能够在正确获得未知对象的聚类中心数目的同时．克服模糊k-均值算法对初始中心点影响的缺陷。将该聚类算法用于确定EBF(椭圆基函数)网络的隐层节点和中心值等参数，在不依赖文本的话者确认实验中．获得了较好的识别效果。

A Modified Clustering Algorithm and Its Application for Speaker Recognition

Fuzzy k-means algorithm based on objective functions among fuzzy clustering algorithms is widely used nowadays. In view of its shortcomings, a new modified algorithm is presented in this paper. With the characteristic of optimization in whole derived from genetic algorithm, this algorithm not only obtained true clustering numbers, but also got the location of centers. With the experiment of text-independent speaker verification, Elliptical basis function neural network with this clustering algorithm can enhance the recognition rate.