Abstract: Based on the speaker vocal feature extracted with independent component analysis(ICA),the Kullback-Leibler(KL) kernel function is derived to develop support vector machine(SVM),which uses KL divergence to show the distance among the features,and the ICA/SVM speaker verification system with high performance is realized.The simulation experiment on speaker verification shows that by using the coefficients of ICA feature basis we obtain a larger margin of separation and fewer support vectors than by training SVM directly on the speech data.The equal error rate(EER) of the system with KL kernel function is lower than those of other traditional SVM methods,so it can be concluded that the SVM method based on KL divergence is highly effective in classification and discrimination.