Deep Neural Networks with Depthwise Separable Convolution for Music Genre Classification
Yunming Liang, Yi Zhou, Tongtang Wan, Xiaofeng Shu · 2019
With the prevalence of internet-based music platforms and the fast increase of music data, considerable attention has been paid to music information retrieval (MIR), like music genre classification. Although the traditional deep neural networks (DNNs) technology for music genre classification has achieved good results, it has a large amount of calculations and parameters due to large data sets and deep networks, resulting in a slow running speed. In an effort to overcome this problem, two types of deep neural networks with depthwise separable convolution are proposed for music genre classification in this paper. The experimental results show the proposed models have better performance on the Extended Ballroom dataset when compared with the traditional models, especially the convolutional neural network with depthwise separable convolution, which achieves an accuracy of 95.10%.