Related Experiment Videos
Adaptive Feature Fusion Gate and Gated Channel-Spatial Attention in CNN-Transformer Models for Music Genre
Yunyan Ma1, Zhenwu Ding1, Shuang Wan1
1School of Music, Jiangxi University of Applied Science and Technology, Nanchang, China.
Plos One
|April 9, 2026
Summary
This study introduces CT-GateNet, a novel deep learning model for automatic music genre classification. It achieves high accuracy on diverse datasets, outperforming traditional methods for music information retrieval.
Area of Science:
- Artificial Intelligence
- Music Information Retrieval
- Machine Learning
Background:
- Automatic music genre classification is vital for managing large music datasets.
- Traditional handcrafted feature methods struggle with scalability.
- Deep learning offers potential for improved music genre classification.
Purpose of the Study:
- To develop an effective deep learning architecture for music genre classification.
- To address data scarcity challenges in audio classification.
- To enhance feature learning and integration for discriminative audio analysis.
Main Methods:
- Proposed Convolutional Neural Network-Gated Transformer Network (CT-GateNet).
- Integrated gated channel-spatial attention and adaptive feature fusion.
- Employed a denoising diffusion probabilistic model for data augmentation.
Main Results:
- Achieved 98.72% accuracy on GTZAN dataset.
- Reached 89.42% and 69.07% on FMA-SMALL and FMA-Medium datasets.
- Demonstrated superior performance and generalization over existing methods.
Conclusions:
- CT-GateNet effectively classifies music genres with high accuracy.
- The hybrid architecture and data augmentation strategy are key to its success.
- Results offer valuable insights for audio classification research.
