Historically, songs have been categorized into genres not just for commercial purposes but also to enhance the listening experience and foster cultural exchange through music. Our primary goal was to compare the performance of traditional machine learning models with more advanced deep learning models like transformers, thereby evaluating the effectiveness of these newer neural network architectures for music genre classification.

We used two datasets, GTZAN and Free Music Archive, training a variety of models on the smaller dataset, and choosing the best performing models to train on the larger. Although it is easy enough to overfit on the training data, creating a model that generalizes well is difficult, especially in the case of imbalanced data.

We further tested our best models in a practical scenario by addressing a contemporary debate: whether Beyoncé’s new album Cowboy Carter is country.