基于多标签流形学习的中国古典音乐情感分析方法研究

项目名称： 基于多标签流形学习的中国古典音乐情感分析方法研究

项目编号： No.61503317

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 刘泱

作者单位： 香港浸会大学深圳研究院

项目金额： 21万元

中文摘要： 音乐情感分析在与情感相关的音乐应用例如音乐推荐、音乐治疗中扮演着重要的角色。目前关于音乐情感分析的研究大多集中于西方音乐，而关于中国古典音乐的情感分析研究相对缺乏。事实上，由于中国古典音乐与西方音乐在传统文化、社会背景、情感倾向性等方面存在着差异，因而在西方音乐体系基础之上所建立的情感模型并不能直接应用到中国古典音乐的研究中来。为了针对中国古典音乐进行情感分析，本项目拟从以下三个方面展开研究：(1)中国古典音乐的情感表示模型；(2)中国古典音乐的特征空间；(3)音乐情感与音乐特征的映射关系。针对以上三个方面，本项目分别提出：(1)基于离散—连续表示的音乐情感模型；(2)基于原始时频信号与等距映射的音乐特征空间；(3)基于多标签流形学习的特征提取算法。本项目拟建立完备的、可量化的分析模型，从而揭示中国古典音乐与其所表达情感之间的内在联系，具有重要的科学意义与现实的应用价值。

中文关键词： 音乐情感分析；中国古典音乐；流形学习；多标签学习；特征提取

英文摘要： Music emotion analysis plays an important role in many emotion-related music applications such as music recommendation and music therapy. However, most of the current researches in this area are focusing on the Western music, there are only few works designed for Chinese classical music. Actually, there are many differences between Chinese classical music and Western music in the aspects of traditional culture, social background, emotional tendentiousness, etc. As a result, the emotion models based on the Western music cannot be applied to Chinese classical music directly. In order to build emotion models for Chinese classical music, in this project, we conduct our research in the following three aspects: (1) emotion representation models of Chinese classical music; (2) feature spaces of Chinese classical music; (3) mapping between the music emotions and music features. Based on the three aspects, our project will propose: (1) the music emotion models based on discrete-continuous representation; (2) the music feature spaces based on the original time-frequency signal and isometric mapping; (3) the feature extraction algorithms based on multi-label manifold learning. This project aims to build a comprehensive and quantifiable analytic model to uncover the intrinsic relationship between Chinese classical music and the emotions it conveys, which has important scientific impact and practical application value.

英文关键词： Music Emotion Analysis;Chinese Classical Music;Manifold Learning;Multi-Label Learning;Feature Extraction