• Chiara Maccani
• Theivan Pasupathipillai
• Carlo Sgorlon Gaiatto

The use of deep learning in Music Genre Classification has become increasingly crucial in recent years, providing powerful tools for managing and categorizing the growing music databases. In this paper, we propose several deep learning models based on Convolutional Neural Networks and Recurrent Neural Networks. We show that using a Convolutional Recurrent Neural Network results in a steeper learning curve, highlighting the efficiency of incorporating Long Short-Term Memory cells to exploit the sequential nature of music. We also experiment with a Multi-Modal architecture to investigate the effect of combining the information conveyed by different audio representations. The performance of the models are evaluated on the Free Music Archive dataset, achieving an overall accuracy of 90% and an F1-score of 60%.