计算机工程与应用 ›› 2007, Vol. 43 ›› Issue (29): 210-213.

• 工程与应用 • 上一篇    下一篇

基于支撑矢量机的汉语方言辨识

顾明亮1,2,3,夏玉果3,张长水1   

  1. 1.清华大学 自动化系,北京 100084
    2.徐州师范大学 物理与电子工程学院,江苏 徐州 221116
    3.江苏省神经与认知工程重点实验室,江苏 徐州 221116
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-10-11 发布日期:2007-10-11
  • 通讯作者: 顾明亮

Support vector machine based Chinese dialect identification

GU Ming-liang1,2,3,XIA Yu-guo3,ZHANG Chang-shui1   

  1. 1.Department of Automation,Tsinghua University,Beijing 100084,China
    2.School of Physics and Electronic Engineering,Xuzhou Normal University,Xuzhou,Jiangsu 221116,China
    3.Jiangsu Key Laboratory of Neural and Cognitive Language Engineering,Xuzhou,Jiangsu 221116,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-10-11 Published:2007-10-11
  • Contact: GU Ming-liang

摘要: 统计学习理论证明,支撑矢量机是具有高分类能力和高推广性能的优秀分类器。但由于语音的动态时间属性,它很难直接应用到汉语方言辨识领域。论文利用高斯混合模型和语言模型提取等维的全局语言特征,成功解决了支撑矢量机难于直接处理动态时间模式的困难,有效地增强了系统的分类能力。实验结果表明,支撑矢量机方法可以比直接用语言模型进行分类决策提高近20%的正确辨识率,比人工神经网络方法也可提高4%的正确辨识率。

Abstract: Statistical learning theory has proved that support vector machine has higher classification ability and higher generalization.However,it is not directly used to Chinese dialect identification,as the speech is a dynamic model.This paper resolves this problem successfully using the global features consisted with the likelihood of Gaussian mixture model and language model,and enhances the discrimination of the system greatly.The experimental results show that SVM based classifier can raise the rate of correct identification about 20% and 4% respectively compared with traditional discriminative classifier and Artificial Neural Network(ANN).