MusicNeXt: Addressing category bias in fused music using musical features and genre-sensitive adjustment layer.

The article focuses on MusicNeXt, a convolutional neural network (CNN) model designed to improve music genre classification (MGC), particularly addressing challenges posed by genre fusion where multiple genres share similar musical features. MusicNeXt enhances feature extraction by fully utilizing temporal information from spectrograms through a depthwise convolutional stacking module and incorporates a genre-sensitive adjustment layer based on L-softmax loss to increase inter-genre distinctiveness and reduce classification bias. Evaluated on three public datasets—GTZAN, ISMIR2004, and Extended Ballroom—MusicNeXt outperforms several state-of-the-art methods in accuracy while maintaining computational efficiency. The study highlights the model's ability to better capture complex fused music characteristics and suggests future exploration of alternative softmax variants for broader applicability in music information retrieval tasks.