论文部分内容阅读
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。
Music category classification includes two stages: feature extraction and classification. Based on the study of wavelet transform theory, the continuous wavelet analysis method is used to extract the characteristic parameters of music. Support vector machines are a classification method specifically for the case of finite samples. It is based on the statistical theory of learning VC dimension theory and the principle of minimum risk based on the sample information based on the model complexity and learning ability to find the best compromise in order to obtain the best promotional capacity. Compared with the traditional mixed Gaussian model and K-nearest neighbor classifier, the classification accuracy is improved by 21% and 23%, respectively, using the exponential radial basis function (ERBF) kernel. Experimental results show that using wavelet and SVM is a very effective method of music classification.