基于示例语义的音乐检索模型  

Music retrieval model based on semantic descriptions

作　　者：秦静[1,2] 林鸿飞[1] 徐博[1] QIN Jing LIN Hong-fei XU Bo(School of Computer Science and Technology, Dalian University of Technology, Dalian 116024, Liaoning, China College of Information Engineering, Dalian University, Dalian 116622, Liaoning, China)

机构地区：[1]大连理工大学计算机科学与技术学院,辽宁大连116024 [2]大连大学信息工程与技术学院,辽宁大连116622

出　　处：《山东大学学报（理学版）》2017年第6期40-48,共9页Journal of Shandong University(Natural Science)

基　　金：国家自然科学基金资助项目(61632011;61562080;61602079)

摘　　要：基于语义描述的音乐检索是根据音乐所表达的语义和对音乐的主观感受,查找或发现音乐的一种方式。一个典型的基于语义描述的检索(query by semantic description,QBSD)系统被定义为有监督的多类别标记(supervised multi-class labeling,SML)模型,通过使用语义相关标签来标记未知,将音乐映射到一个"语义空间",从而克服语义鸿沟问题。在SML模型基础上,提出将示例音乐作为检索条件,通过对音乐语义的标注将检索示例映射到语义空间,然后在标记后的数据库中,返回语义相似的音乐。并且采用深度学习算法,设计了多类别标记模型。实验表明该模型能够满足用户基于语义音乐检索的基本需要。Query by semantic description（QBSD）is a natural way to retrieve and discovery relevant music based on semantic contents and users’subjective feelings.A QBSD system can be defined as a supervised multi-class labeling bridging（SML）model semantic for the semantic gaps,by which a song could be tagged using semantic labels on and mapped into spaces.song In this paper,we propose a method could for be querying used by semantic description retrieved based the SML model,in which The a represented song as a semantic most vector as a query,and within the tagged network music dataset.resulted into list contains similar songs show in that the semantic space.A convolutional could neural is of also integrated the SML model.space The experiments the proposed method obtain relevant pieces music in the same semantic effectively and efficiently.

关 键 词：语义描述检索 有监督多类别标记 卷积神经网络 音乐检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]