QQ客服 官方微博 反馈留言

	音乐情感分类中关键问题的研究
	王愈 ; 蔡锐 ; 蔡莲红 ; Yu Wang ; Rui Cai ; Lian-Hong Cai
	2010-07-15 ; 2010-07-15
会议名称	第一届建立和谐人机环境联合学术会议（HHME2005）论文集 ; 第一届建立和谐人机环境联合学术会议（HHME2005） ; 中国昆明 ; CNKI ; 中国计算机学会、中国图象图形学学会、ACM SIGCHI中国分会、清华大学计算机科学与技术系
关键词	音乐情感分类 情感段预切分 混淆度 music mood classification, pre-segment, confusion TP391.42
其他题名	Key Problems of Music Mood Classification
中文摘要	随着基于内容的音乐检索的研究不断深入,音乐情感分类的重要性也逐渐凸现出来。本文基于一个音乐情感分类系统,对其中的“情感段预切分”和“声学特征在情感分类中的适用度”等问题进行了较为深入的探讨和研究,提出了一种切分情感段的方法,并使用“混淆度”等指标对特征的适用度进行了量化的评价。基于西方古典音乐的实验结果表明,该方法在大多数实验数据上能够较为准确地界定情感边界。; Music is a carrier of mood; mood is the connotation of music. To detect and classify mood of music is a new domain of multimedia content management. Nowadays, music mood clasmposed of three majsification is generally used for content-based retrieval of music, which gains more and more attention. In this paper, we focus on two key problems of music mood classification: pre-segment and applicability of acoustic features for classification, based on a music mood classification system. On mood taxonomy, we select Thayer’s two-dimensional mood model. Referring to Dan Liu’s system (Dan Liu, Lie Lu, Hong-Jiang Zhang, 2003), the structure of our system here, which is designed according to the flow of data, is coor modules: frame-level analysis, clip-level analysis and classifier decision. In frame-level analysis model, we extract 27 features for each frame which could be clustered into three groups: temporal features, spectral shape features and spectral contrast features. In clip-level analysis module, we first segments music into clips with consistent mood approximately through pre-segment procedure, and then extracts 57 features, including: mean and standard deviation of the frame-level features, low energy and rhythm features. Classifier decision module uses GMM as its classification model. Usually the mood in a song is not consistent. Pre-segment deals with this problem, by way of segmenting the song into clips with consistent mood approximately. It is discussed detailedly in this paper, since it is significant to the capability of the system and is new in a way. According to some knowledge about music, we infer that this problem could be transformed into the segment of music paragraph or movement. The segment bases on K-L distance, using the frame-level features. There are four key procedures: removing silent part, selection and organization of features, amalgamation of features` decision, combination of ill-short clips。Experiments on western classic music indicate that the proposed algorithm can locate the boundary of mood exactly. Applicability of acoustic features for classification denotes the capability of distinguishing two types of mood. To describe this capability quantitatively, we propose a concept named confusion, which is defined as the probability of failure in the case of classifying two types of mood.; 国家自然科学基金重点项目(60433030)
语种	中文 ; 中文
内容类型	会议论文
源URL	[http://hdl.handle.net/123456789/70012]
专题	清华大学
推荐引用方式 GB/T 7714	王愈,蔡锐,蔡莲红,等. 音乐情感分类中关键问题的研究[C]. 见:第一届建立和谐人机环境联合学术会议（HHME2005）论文集, 第一届建立和谐人机环境联合学术会议（HHME2005）, 中国昆明, CNKI, 中国计算机学会、中国图象图形学学会、ACM SIGCHI中国分会、清华大学计算机科学与技术系.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。