We proposed a novel approach that is based on the kernel K-means clustering and support vector machine （SVM）

set up a support vector machine for each two speakers

here kernel K-means clustering is exploited to input speech signal of SVM into a given amount of clusters choosing the effective samples as the input of SVM. In this paper

we proposed approach acquiring the effective of clustering and reduction of speech data

and it greatly improved recognition rate. The experimental results show the proposed approach performances are better than tradition approaches in the SVM as classier.