摘要
多版本音乐识别,作为音乐信息检索领域内的一个重要课题,得到了人们的广泛专注.受海量音乐资源的驱动,人们对高效算法的需求呼之欲出.词袋模型是一种在自然语言处理和数字图像处理领域中的广泛应用的特征表示模型,本文将其引入到多版本音乐识别技术研究中,基于目前业界广泛应用的Chroma特征,提出一种新的高层次特征,Chroma-based BOW特征对歌曲的旋律信息进行表示,实现了特征空间到线性矢量的降维映射.同时,设计并实现了一套完整的多版本音乐识别算法.实验证明该特征对于多版本音乐识别是有效的,相应的识别算法能够大幅度提高了传统音乐识别系统的效率.
Cover song detection, as an important topic in music information retrieval,has attracted widespread focus. Driven by the big data of music resources, an efficient algorithm becomes urgent. BOW is a feature expression model that widely used in natural lan- guage processing and digital image processing. In this paper,we apply this model to cover song detection,extracting Chroma-based BOW feature to express songs＇ melody information based on widely used Chroma in this field. This makes the map from the feature space to the linear vector space available by dimension reduction. We also design and implement a series of algorithms of cover song detection. The feature is found to be useful in the cover song detection experiments. The corresponding recognition algorithm can great- ly improve the efficiency of traditional music information retrieval system.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第2期397-400,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61171128)资助
国家"八六三"计划项目(2011AA01A109)资助
