音乐作品通常具有多个不同版本.重新演奏、演绎使得各版本音乐在音质、速度、节奏、结构、基调、和声、歌词等多个方面都可能产生不同,这使得多版本音乐识别成为一个具有挑战性的研究领域.近年来多版本音乐识别技术发展迅速,本文对其包含的五个关键技术模块进行描述,分别是特征提取模块、基调不变性模块、速度不变性模块、结构不变性模块、相似性计算模块;其次概要介绍该领域的典型算法;最后总结当前存在的问题,并讨论进一步的研究方向.
There usually exist various versions(covers) of a music work.New performances or interpretations of a music piece may cause differences on timbre,tempo,rhythm,structure,key,harmony and lyrics etc.,which makes cover song identification a challenging task.In this paper,we summarize five key technical modules which are feature extraction,key invariance,tempo invariance,structure invariance and similarity computing,then introduce representative algorithms,at last discuss the unsolved issues and future directions.