基于支撑向量机的说话人确认系统

Speaker Verification System Based on Support Vector Machines

支撑向量机(SVM)是一种新的统计学习方法,和以往的学习方法不同的是SVM的学习原则是使结构风险(Structural Risk)最小,而经典的学习方法遵循经验风险(Empirical Risk)最小原则,这使得SVM具有较好的总体性能.文章提出一种基于支撑向量机的文本无关的说话人确认系统,实验表明同基于向量量化(VQ)和高斯混合模式(GMM)的经典方法相比,基于SVM的方法具有更高的区分力和更好的总体性能.

Support Vector Machine (SVM)is a new statistical learning methods．Compared with other machine learning methods,the learning discipline of SVMs is to minimize the structural risk instead of empirical risk the learning discipline of classical methods,and it gives SVMs better generative performance．This paper proposes a text－independent speaker verification system based on support vector machines．The experiments show that performance of the system based on SVMs is better than those systems based on VQ or GMM．