音乐情感分类系统技术研究及应用
发布时间：2018-12-10 13:17
【摘要】： 随着当今我国社会信息化的迅速发展,各种多媒体信息资料以飞快的速度不断产生。为了对这些不断激增的多媒体资料进行有效地存储、交流、共享,对多媒体信息资料的自动索引和检索技术提出了一系列新的要求。本文就其中的音乐情感自动分类技术作了深入的研究。 目前,商用的音乐检索系统基本都仅限于文本检索技术为主,而信息的输入则以手工为主。这些系统难以对飞速产生的音乐资料进行及时有效地索引和检索。一方面,仅仅基于文本的信息已经越来越难于满足用户检索的需要;另一方面,手工标注的效率无法及时索引浩如烟海的音乐作品。因此,如何采用新的技术手段来满足这些要求,对索引和检索技术提出了新的课题。 本文针对上述音乐索引和检索技术中出现的新特点,提出了采用基于内容的音乐情感分类检索技术。对现存的一些基于内容的音乐情感分类检索技术进行了分析和比较,根据国内广电行业的技术要求,基于GMM音乐情感分类系统框架和技术,设计了一个音乐情感自动分类系统,并针对现有分类器“大一统”的不合理性,通过大量试验分析,确定各音频特征在分类过程中的适用度,提出了对分类器的层次化改进方法,一定程度的提高了分类系统的准确度。同时,对音乐如何进行基于情感的“预分段”问题进行了研究,采用K-L距离来计算相邻单位段间的差异性,将整首歌曲分割为若干具有一致情感的音乐片段,从而为更好的进行音乐情感分类奠定了基础。 本文提出的音乐情感自动分类检索技术还在一个面向上海文广新闻传媒集团的音频资产管理系统中进行了实现,并在实际使用中取得了较好的效果,为音乐情感的进一步分析采集积累了宝贵数据。
[Abstract]:With the rapid development of information technology in our country, various multimedia information materials are produced at a fast speed. In order to store, communicate and share these multimedia data effectively, a series of new requirements are put forward for the automatic indexing and retrieval of multimedia information. This paper makes a deep research on the automatic classification of music emotion. At present, the commercial music retrieval system is mainly limited to text retrieval technology, while the input of information is mainly manual. It is difficult for these systems to index and retrieve the rapidly generated music data in a timely and effective manner. On the one hand, text-based information has become more and more difficult to meet the needs of users to retrieve; on the other hand, the efficiency of manual tagging can not index a vast number of music works in time. Therefore, how to adopt new technical means to meet these requirements, and put forward a new subject of indexing and retrieval technology. In view of the new features of the music indexing and retrieval techniques mentioned above, a content-based music emotion classification retrieval technique is proposed in this paper. This paper analyzes and compares some existing content-based music emotion classification and retrieval technologies. According to the technical requirements of domestic radio and television industry, based on the framework and technology of GMM music emotion classification system, an automatic music emotion classification system is designed. Aiming at the irrationality of the existing classifier, this paper determines the applicability of each audio feature in the process of classification through a large number of experiments and analyses, and puts forward a hierarchical improvement method for the classifier. The accuracy of the classification system is improved to a certain extent. At the same time, how to carry out the "pre-segmentation" based on emotion in music is studied. K-L distance is used to calculate the difference between adjacent units, and the whole song is divided into several music fragments with consistent emotion. Therefore, it lays a foundation for better classification of music emotion. The automatic classification and retrieval technology of music emotion proposed in this paper is also implemented in an audio asset management system for Shanghai Wenguang News Media Group, and has achieved good results in practical use. It accumulates valuable data for the further analysis of music emotion.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2006
【分类号】：J618.9
本文编号：2370636
[Abstract]:With the rapid development of information technology in our country, various multimedia information materials are produced at a fast speed. In order to store, communicate and share these multimedia data effectively, a series of new requirements are put forward for the automatic indexing and retrieval of multimedia information. This paper makes a deep research on the automatic classification of music emotion. At present, the commercial music retrieval system is mainly limited to text retrieval technology, while the input of information is mainly manual. It is difficult for these systems to index and retrieve the rapidly generated music data in a timely and effective manner. On the one hand, text-based information has become more and more difficult to meet the needs of users to retrieve; on the other hand, the efficiency of manual tagging can not index a vast number of music works in time. Therefore, how to adopt new technical means to meet these requirements, and put forward a new subject of indexing and retrieval technology. In view of the new features of the music indexing and retrieval techniques mentioned above, a content-based music emotion classification retrieval technique is proposed in this paper. This paper analyzes and compares some existing content-based music emotion classification and retrieval technologies. According to the technical requirements of domestic radio and television industry, based on the framework and technology of GMM music emotion classification system, an automatic music emotion classification system is designed. Aiming at the irrationality of the existing classifier, this paper determines the applicability of each audio feature in the process of classification through a large number of experiments and analyses, and puts forward a hierarchical improvement method for the classifier. The accuracy of the classification system is improved to a certain extent. At the same time, how to carry out the "pre-segmentation" based on emotion in music is studied. K-L distance is used to calculate the difference between adjacent units, and the whole song is divided into several music fragments with consistent emotion. Therefore, it lays a foundation for better classification of music emotion. The automatic classification and retrieval technology of music emotion proposed in this paper is also implemented in an audio asset management system for Shanghai Wenguang News Media Group, and has achieved good results in practical use. It accumulates valuable data for the further analysis of music emotion.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2006
【分类号】：J618.9
【引证文献】
相关硕士学位论文 前4条
1 李静;基于情感标签的音乐检索算法研究[D];大连理工大学;2010年
2 孙向琨;音乐内容和歌词相结合的歌曲情感分类方法研究[D];苏州大学;2011年
3 孙坚;基于语义的音乐检索方法研究及试验[D];重庆大学;2010年
4 高友平;数字化音乐资源情感检索技术研究[D];华中师范大学;2013年
，本文编号：2370636
本文链接：https://www.wllwen.com/wenyilunwen/qiyueyz/2370636.html
