Music Genre Classification Using Attention-Based CNN-Feature Fusion Paradigm

In order to organise the enormous music collections available on the Internet, genre hierarchies are widely used. Automatic musical genre classification, which may supplement or even replace the human user in this process, would substantially assist music information retrieval systems. A framework for developing and evaluating features for any content-based analysis of musical signals is also provided by automatically categorising musical genres. The paper's primary goal is to demonstrate how audio signals may be automatically classified using two models into a hierarchy of musical genres. Two feature sets are recommended for expressing timbral texture and rhythmic content, respectively. The CNN model with attention, which uses the fusion of timbral and rhythmic features, has the highest classification accuracy of 85%.