摘要
音乐类型分类主要包括两个阶段:特征提取和分类。文中在研究小波变换理论基础上,采用连续小波分析方法提取音乐特征参数。支持向量机是专门针对有限样本情况下的一种分类方法。它是建立在统计学习理论的VC维理论和结构风险最小原理基础上,根据有限的样本信息在模型的复杂性和学习能力之间寻求最佳折衷,以期获得最好的推广能力。采用指数径向基函数(ERBF)内核,分类正确率可达85%,比传统的混合高斯模型和K近邻分类器,分类性能分别提高了21%和23%。实验结果表明,采用小波和支持向量机方法是一种相当有效的音乐类型分类方法。
Musical genre classification task falls into two major stages： feature extraction and classification. According to a research in wavelet theory, continuous wavelet analysis method is used to extract feature parameters of music. SVM is designed to classifying of limited samples. It is based on VC dimension and the ERM（expectation risk minimum）of statistical learning theory. According to information of limited samples, there is a trade- off existing between models complexities and learning capability to get best extending ability. Exponential radial basis function （ERBF） kernel function are used to classify the musical genre,85% of classification are correct. In comparison with Gaussian mixture model （GMM） classifier and K nearest neighboring （KNN） classifier, the classification performances are improved by 21% and 23% respectively. Experimental results indicate that wavelet and SVM is useful, method for musical genre classification.
出处
《计算机技术与发展》
2008年第12期19-21,24,共4页
Computer Technology and Development
关键词
音乐类型分类
小波
支持向量机
核函数
musical genre classification
wavelet
support vector machines
kernel function
