Music genre classification with parallel convolutional neural networks and capuchin search algorithm.

Area of Science:

• Computer Science
• Music Information Retrieval
• Machine Learning

Background:

• Music genre classification is crucial for personalized music experiences and playlist generation.
• High-tech multimedia tools necessitate robust music classifiers for efficient handling of unlabeled music data.
• Machine learning and deep learning approaches are essential for developing advanced music classification systems.

Purpose of the Study:

• To develop a novel ensemble system of convolutional neural network (CNN) models for accurate music genre detection.
• To enhance music classification by integrating diverse audio features and optimizing model hyperparameters.
• To improve consumer experiences with media players and music files through better genre categorization.

Main Methods:

• Utilized an ensemble of convolutional neural network (CNN) models for music genre classification.
• Employed discrete wavelet transform (DWT), mel frequency cepstral coefficients (MFCC), and short-time Fourier transform (STFT) for comprehensive feature extraction.
• Optimized model hyperparameters using the capuchin search algorithm (CapSA).

Main Results:

• Achieved an average classification accuracy of 96.07% on the GTZAN dataset.
• Achieved an average classification accuracy of 96.20% on the Extended-Ballroom dataset.
• Demonstrated superior performance compared to previous comparable methods through integrated signal processing and CNN models.

Conclusions:

• The proposed ensemble CNN method effectively classifies music genres by combining multiple signal processing techniques.
• The integration of DWT, MFCC, and STFT features provides a robust framework for capturing stylistic musical qualities.
• This research advances music genre classification accuracy and offers insights into blending diverse musical components.