用于说话人识别支持向量机模型的核函数选择

摘要

支持向量机(SVM)已广泛地应用于文本无关的说话人辨认系统,不同的核函数影响识别性能.基于此,在TIMIT语料库上对线性核、多项式核以及径向基核进行了对比实验.实验表明多项式核在多项式次数等于6的情况下具有最佳的识别性能,其识别率可以达到82.88％.%Support vector machine (SVM) has been widely used in text-independent speaker recognition system, different kernel functions sway the performance of recognition. Based on this, we conduct the contrast experiments on linear kernel, polynomial kernel and RBF kernel on TIMIT corpus. The experiments show that polynomial kernel has best recognition performance when with a degree of 6, and the recognition rate can reach up to 82. 88% .