摘要
研究一种用支持向量机(SVM)进行多类音频分类的方法,其中引入增广两类分类法(AB法)设计多类分类器。该算法把音频分为四类:音乐、纯语音、带背景音的语音和典型的环境音,并分析了这几类音频的八个区别性特征,包括修正低能量成分比率(MLER)和修正基频(MPF)两个新特征以及频域总能量、子带能量、频率中心等其它六个基本特征,综合考察了不同特征集在基于SVM分类器中的分类精度。实验结果表明,提取的音频特征有效,基于SVM的多类音频分类效果良好。
In this paper we studied a multi-class audio classification algorithm performed by the support vector machine(SVM),in which augmented binary-class(AB) classification method was introduced to design the multi-class classifier.Four classes were considered in audio frequency:music,pure speech,speech over background and typical environment sound,and eight discriminating features of these audio classes were analysed,including the two new ones proposed by the author-modified low energy component ratio(MLER) and modified pitch frequency(MPF),and other six basic features such as spectrum energy,sub-band powers,frequency centroid,etc.The classification accuracies of different features' set in SVM-based audio classifier were evaluated comprehensively.Experiment results show that the audio features selected in this paper are effective for audio classification,and the result of applying SVM to multi-class audio classification is good.
出处
《计算机应用与软件》
CSCD
2010年第4期98-101,共4页
Computer Applications and Software
基金
上海高校选拔培养优秀青年教师科研专项基金项目(DXZ06007)
关键词
支持向量机
音频分类
增广两类分类法
Support vector machine Audio classification Augmented binary-class classification method