Genre Based Music Classification using Machine Learning and Convolutional Neural Networks

In this digital era, as the quantity of music being released on a regular basis continues to grow, Music Information Retrieval (MIR) has become a challenging yet important task. Music Classification is a crucial part of MIR. This involves the task of defining and implementing measurements of music similarity, which is quite complex. Pattern recognition and machine learning techniques offer appropriate solutions to deal with such issues. In this paper, we have built various machine learning models for genre classification and trained them on the GTZAN dataset. We have used signal processing to extract informative features from raw audio and have analyzed them for usage. Along with traditional models, we have used ensemble tree models, which result in higher accuracy. A deep learning approach is also taken using convolutional neural networks for the classification task. The proposed CNN model achieves the best performance among all. We have compared the models using various metrics and logged their results along with suitable graphs.