A novel music genre classification algorithm based on Continuous Wavelet Transform and Convolution Neural Network

Abstract

The automated genre categorization of musical audio signals plays a fundamental role in the application, spectral characteristics that have been averaged across a large number of audio frames are estimated by Continuous Wavelet Transform (CWT) and then converted to a gray-scale picture for training and classification. A Convolution Neural Networks (CNN) model is proposed in this paper and the results demonstrated that the classification accuracy of the proposed CWT+ CNN model is about 70% with the testing sample and 94% with the training sample. This may suggest good application potential for the music genre classification.