Computer Engineering and Applications ›› 2017, Vol. 53 ›› Issue (16): 138-143.DOI: 10.3778/j.issn.1002-8331.1603-0441

Previous Articles     Next Articles

Music classification based on Hidden Markov Models

XIAO Xiaohong1, ZHANG Yi2, LIU Dongsheng1, OUYANG Chunjuan1   

  1. 1.School of Electronics and Information Engineering, Jinggangshan University, Ji’an, Jiangxi 343009, China
    2.Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
  • Online:2017-08-15 Published:2017-08-31

基于隐马尔可夫模型的音乐分类

肖晓红1,张  懿2,刘冬生1,欧阳春娟1   

  1. 1.井冈山大学 电子与信息工程学院,江西 吉安 343009
    2.清华大学 电子工程系,北京 100084

Abstract: Music genre is one of the most common ways used in digital music database management. A music automatic classification scheme based on Hidden Markov Models (HMMs) is proposed. While considering traditional timbre, another important feature--Tempo is taken into consideration. Meanwhile, the bagging is used to train two groups of HMM for classification, which obtain good results. In this paper, optimized hyper-parameters in the HMM structure, the number of states and Gaussian mixtures, and find the best HMM parameters. Furthermore, the traditional model and original model are tested on the well-known GTZAN database. The results show that the proposed method considering tempo feature acquires better classification accuracy compared to the traditional model.

Key words: classification, music genre, tempo, Hidden Markov Models

摘要: 音乐类型(Genre)是应用最普遍的管理数字音乐数据库的方式,提出一种基于隐马尔可夫模型(Hidden Markov Models,HMMs)的音乐自动分类方案。在考虑传统的音色特征(Timbre)的同时,将另一重要特征节奏(Tempo)也加以考虑,并通过bagging训练两组HMM进行分类,达到了良好的效果。从结构、状态数和混合高斯模型数三个方面进行了参数优化,找到了最佳的HMM参数。在音乐数据集GTZAN上对传统模型和新模型分类效果进行了测试,结果表明考虑了节奏特征的HMM分类效果更佳。

关键词: 分类, 音乐类型, 节奏, 隐马尔可夫模型