基于内容的音乐节拍跟踪  

Content Based Music Beat Tracking

作　　者：陈哲[1] 许洁萍[1] 

机构地区：[1]中国人民大学信息学院计算机系,北京100872

出　　处：《电子学报》2009年第B04期156-160,共5页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60873098);中国人民大学"985工程"自由探索项目(No.21356234)

摘　　要：节奏是音乐的三大要素之一,对其准确的分析和提取具有重要的研究意义.节奏特征主要分为音乐节拍和速度.本文首先提出了一种利用自相关相位-熵序列分析音乐节拍结构及音乐速度的方法.利用该方法对50首流行歌曲及50首纯乐器音乐速度的分析结果可达到97%;在速度分析结果基础上,文中还给出了节拍点求解过程的近似贝叶斯模型,使得节拍点序列在整体上与音乐信号的长时速度保持一致;文中在最后给出了利用动态规划思想进行音乐节拍跟踪的新方法,完成了音乐节拍跟踪实验,并通过与其它实验的结果比较,验证了算法的有效性.Rhythm is one of the three main factors of music. It＇s of great importance to analyze and extract rhythm features from music signal which mainly includes beat and tempo. In this paper, firstly we introduce a method which uses autocorrelation phase-entropy,to analyze meter and tempo of music. With this method we get a 97 percent of accuracy degree in music tempo induction on a database of 100 songs（including 50 popular and 50 Chinese folk） ; Then based on the result of tempo induction, we propose a roughly Bayesian model to keep the beats sequence consistent with the global tempo of music in the music beats locating;Finally we introduce a new beat tracking algorithm using dynamic programming to finish the teat tracking. By comparing its result with others＇, this algorithm proves to be very effective.

关 键 词：节奏特征 节拍跟踪 速度提取 自相关-熵序列 动态规划 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]