基于听觉图像的音乐流派自动分类

音乐流派的自动分类是音乐信息检索系统的重要组成部分.将听觉图像引入音乐流派的分类研究中,用听觉图像模型模拟人耳耳蜗结构,基于音乐流派分类研究常用的GTZAN数据库,将一维音频信号转换为二维听觉图像,对音乐听觉图像进行尺度不变特征转换（SIFT）及空间金字塔匹配（SPM）,从局部到整体地提取图像的纹理特征,最后采用LibSVM中线性核函数的支持向量机对音乐流派进行分类.实验结果表明,与同样基于人耳耳蜗结构提出的美尔频率倒谱系数（MFCC）流派分类方法相比,基于听觉图像的流派分类正确率提高15%.

Automatic Music Genre Classification Based on Auditory Images

Automatic music genre classification is an important part of the music information retrieval system. The concept of ＂auditory image＂ is introduced into music genre classification in this paper. Auditory image model （AIM）converts the one-dimensional audio signal into two-dimensional auditory images by simulating the hu- man ear cochlear structures for the commonly database of GTZAN. And then, the methods of scale invariant feature transformation （SIFT） and space pyramid matching （SPM） are used to extract image features from the part to the whole. And the linear kernel support vector machine is chosen for classification since the dimension of features was high. Experimental results show that the genre classification accuracy based on the auditory images can be 15% higher than the Mel-frequency cepstral coefficients （MFCC） which is also based on the cochlear structure of the human ear.