基于CGABC-SVM的多特征融合音乐分类算法研究

Multi Feature Fusion Music Classification Algorithm Based on CGABC-SVM

对电子音乐进行合理且有效的分类,可以使用户能快速搜索到喜爱的音乐,也使音乐推荐系统能进行更加精准的推荐。为了提高音乐分类的准确性,论文提出了基于CGABC-SVM的多特征融合音乐分类方法。在特征提取方面,针对单一音频特征表达不完整的问题,提取基音频率、共振峰、梅尔频率倒谱系数和相对谱-感知线性预测4种音频特征,组成多特征融合矩阵。在分类器选择方面,针对支持向量机(SVM)参数难以选取的问题,论文使用交叉全局人工蜂群算法(CGABC)来优化SVM的参数,构建CGABC-SVM音乐分类模型。实验结果表明,论文音乐分类方法可以有效地区分各种音乐信号,音乐分类的准确性显著好于对比音乐分类方法。

Reasonable and effective classification of electronic music can make users quickly search for their favorite music and make music recommendation system more accurate.In order to improve the accuracy of music classification,this paper proposes a multi feature fusion music classification method based on CGABC-SVM.In the aspect of feature extraction,aiming at the problem of incomplete expression of single audio feature,four kinds of audio features,including pitch frequency,formant,Mel frequency cepstrum coefficient and relative spectrum perception linear prediction,are extracted to form a multi feature fusion matrix.In terms of classifier selection,aiming at the problem that the parameters of support vector machine(SVM)are difficult to select,this paper uses the cross global artificial bee colony algorithm(CGABC)to optimize the parameters of SVM,and constructs the CGABC-SVM music classification model.The experimental results show that the music classification method can effectively distinguish various music signals,and the accuracy of music classification is significantly better than that of comparative music classification method.