一种基于分层结构的音乐自动分类方法

New Music Genre Classification Method via Hierarchical Support Vector Machines

音乐分类在音乐信息检索中占据重要地位,自动的音乐分类方法可以在降低花费的同时取得良好的精准度.传统的平面型音乐流派分类方法主要从全体数据集出发使用若干特征作为分类依据,导致分类效果并不太好.本文考虑了音乐文件本身的属性,结合统计学方面的特征,提出了基于分层结构的分类方法.该方法首先使用K-Means聚类方法以分析不同类别间的关系,并构造类别层次关系图,在此基础上使用支持向量机方法进行分类,通过使用不同的特征集合,保证了分类的准确率.该方法在GTZAN数据集上进行了相关实验,实验结果表明本文所提出的方法能够取得较好的分类准确率.

Music genre classification plays an important role in music information retrieval. Automatic methods for music genre classification can save lots of money while achieve the affordable precision. Conventional flat music genre classification methods pay more attention to the whole dataset with many features,so the precision of the classification is not so satisfied. This paper paid more attention on the music features combining with statistical features,and proposed a hierarchical structure for the music classification. At first,KMeans clustering is used to analysis the relations among all the genres,and the hierarchical structure is built. Then support vector machine is used to classify all the genres into different groups with different feature sets,which assures the precision. Some experiments are implemented with GTZAN dataset,and the results show that the proposed method can achieve good classification results.