Music Genre Recognition with Handcrafted Audio Features
André Luís de Oliveira, Luiz F. Carvalho, Daniel Prado Campos, Rafael Gomes Mantovani · 2024
Music has been increasingly prominent in people’s lives in recent years. Technology integration with music is progressively increasing, directly contributing to the enhancement and understanding of this art form. In this sense, Music Genre Recognition (MGR) applies Machine Learning (ML) to identify similar patterns in music and classify them. This study assessed traditional ML algorithms in the GTZAN dataset, classifying audio songs into ten genres. The results showed that the XGBoost classifier statistically outperformed all the other algorithms evaluated, with an accuracy value of 0.722. Future experiments can improve it with a more robust feature engineering process, exploring and mixing deep with traditional features.