改进K-means算法在声纹识别中的应用

在基于高斯混合模型（GMM）的声纹识别算法中,K-means聚类算法是GMM模型参数初始化常用的方法之一。传统K-means算法在聚类过程中采用几何距离进行分类,忽略了类中各矢量的分布不同对聚类结果的影响,常常得不到令人满意的识别结果。文中对传统K-means算法进行了改进,并将改进后的K-means算法与GMM结合应用到声纹识别系统中。实验结果表明,改进的K-means算法与传统的算法相比具有更好的识别效果。

Application of Improved K-means Algorithm to Voice Print Recognition

In voice print recognition algorithm based on Gaussian mixture model（GMM）,the algorithm of K-means clustering is one of widely used methods in GMM.Traditional K-means algorithm was classified by geometrical distance in clustering process and ignored the influence of different distribution in category for clustering results,therefrom,the recognition results were usually suboptimal.The traditional K-means algorithm was improved in the thesis.The improved algorithms with GMM were all applied to voice print recognition.Experimental results expatiates that the improved K-meams algorithm has better recognition performance than that of the traditional one.