基于局域判别基的音频信号特征提取方法  

AUDIO SIGNAL FEATURE EXTRACTION METHOD BASED ON LOCAL DISCRIMINANT BASIS

作　　者：魏国华[1] 郑继明[2] 

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065 [2]重庆邮电大学应用数学研究所,重庆400065

出　　处：《计算机应用与软件》2009年第11期84-87,共4页Computer Applications and Software

基　　金：重庆市教委科学技术项目(KJ080524)。

摘　　要：音频特征提取在音频信号分析和处理中起着非常重要的作用。考虑到音频信号的非平稳性,对音频信号进行小波包分解,为了获取健壮的特征,采用改进的局域判别基(LDB)技术对小波包树进行裁剪,提取局域判别基各子空间能量的统计特征组成特征矢量,并利用Fisher准则函数进行特征选择,根据特征矢量设计支持向量机分类器,对三类音频进行分类识别。实验结果表明,该方法提取的特征矢量在音频信号分类中是非常有效的。Audio feature extraction plays an important role in analyzing and processing audio signals. Wavelet packet decomposition is performed considering the non-stationary of audio signals. In order to get robust features ,the modified LDB technique is used to prune the wavelet packet tree, the statistical features of energy are extracted from each sub - space of the LDB to form feature vector, and feature choice is done with Fisher criterion function. A support vector machine classifier is designed according to the feature vector for classification and recognition of three audio classes. The results of experiment show that the feature vector extracted is very efficient in audio signal classification.

关 键 词：特征提取 局域判别基 FISHER准则 支持向量机 

分 类 号：TN912.3[电子电信—通信与信息系统] TN911.7[电子电信—信息与通信工程]