基于KL散度的支持向量机方法及应用研究

The SVM Method Based on KL Divergence and Its Application

针对ICA提取的说话人语音特征,导出以库尔贝克—莱布勒(KL)散度作为距离测度的KL核函数用来设计支持向量机,实现了一个高分辨率的ICA/SVM说话人确认系统.说话人确认的仿真实验结果表明,使用ICA特征基函数系数比直接使用语音数据训练SVM得到的分类间隔大,支持向量少,而且使用KL核函数的ICA/SVM系统确认的等差率也低于其它传统SVM方法,证明了基于KL散度的支持向量机方法在实现分类和判决上具有高效性能.*

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 throe 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.