一种新的高效大型音乐数据库查询算法  

A new effective query method for large music databases

作　　者：刘怡[1] 郝云飞[1] 

机构地区：[1]中国人民大学数据工程与知识工程研究所,北京100872

出　　处：《湖南科技大学学报（自然科学版）》2006年第1期68-72,共5页Journal of Hunan University of Science And Technology：Natural Science Edition

摘　　要：在基于哼唱的音乐信息检索领域中,查询处理算法一直是一个重要的研究课题。根据人对乐曲相似理解的实验结果,设计了新的基于单侧连续匹配的可容错的近似匹配的算法.此算法采用n-gram的顺序hash索引,加快了可容错的查询处理速度.还设计了一系列包括各种类型和各种数量错误的查询的实验,来比较该算法和其他两种常用算法的查询处理速度和容错能力.当测试数据集为78000首网上收集的MIDI乐曲段,包含各种错误的查询为1000个时,该算法取得79%的前3位命中率和87%的前10位命中率.这些结果表明,该方法的性能高于现有的算法.表2。This paper introduces a new effective and tolerance query method for large music databases. According to the findings of a series experiments for music similarity of human perceiving, a new approximate matching algorithm based on one side consecutive match is designed. The algorithm uses a sequence hash index with n-gram to speed the fault tolerance query processing. A series of experiments including input set with various kinds and amount errors have been carried out. With the support of the experiment results, the performance of query processing and capability of fault tolerance of our method and other two approaches are compared. The experiments data show the efficiency of our method. When test datasets are 78 000 MIDI pieces and query sets including various errors are 1 000 phrases,the top-3 success rate of our method is 79% and the top-10 success rate is 87%. 2tabs., 8refs.

关 键 词：基于内容的音乐检索 n-gram方法 近似匹配 查询扩展 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]