特征组合的中文音乐情感识别研究

Music emotion recognition of different features combinations based on SVM

音乐情感识别是音乐检索的一个重要组成部分.基于音乐声学特征分析,尝试提取代表音乐声学特性的时域、频域、倒谱域的各种特征,并利用支持向量机(support vector machine,简称SVM)算法对中文音频进行情感分类,以研究不同特征组合在音乐情感分类上的效果.通过对比各种不同特征组合的音乐情感识别效果,发现由4个时域特征、频谱、幅度谱和相位谱组成的音乐特征对中文音乐情感分类的效果良好.

In this paper,SVM was used to classify music emotions by analyzing music features and comparing recognition results of combination of different acoustic characteristics of music.First,time domain,frequency domain,cepstral domain features of music were extracted and these features were combined to form the training samples,then SVM was chosen to build the classifier which can classify music into different emotional categories.By comparing each combination,we find that the feature combination which consists of four time domain features,frequency spectrum feature and magnitude spectrum feature and phase spectrum feature showed is more effective.