Music Genre Classification using Deep Neural Networks

• Published in: 2023 35th Chinese Control and Decision Conference (CCDC) pp. 2384 - 2391
• Main Authors: Yimer, Mekonen Hiwot, Yu, Yongbin, Adu, Kwabena, Favour, Ekong, Liyih, Sinishaw Melikamu, Patamia, Rutherford Agbeshi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20-05-2023
• Subjects: classification; Convolution; Convolutional neural networks; Feature extraction; Image segmentation; Mel frequency cepstral coefficient; Mel Frequency Cepstral Coefficients; Music genre; Time-frequency analysis; Training
• ISSN: 1948-9447
• DOI: 10.1109/CCDC58219.2023.10327367

Classifying music to its genre is one of the most challenging tasks in Music Information Retrieval (MIR). Music genre classification has been a critical activity in recent years due to the increasing development of online and offline music tracks. To make these tracks more accessible, they need to be indexed correctly. This paper reviews the current state-of-the-art methods in music genre classification and proposes a new approach using the Deep Convolution Neural Network (DCNN) model. To extract feature vectors and classify music into their respective genres, two models were designed, implemented, and evaluated on the Mel Frequency Cepstral Coefficients (MFCCs) of the songs: a 16-layered Convolutional Neural Network (CNN) named Music Genre Convolutional Neural Network (MG-CNN) and a pre-trained Deep Neural Network (DNN) VGG16 named Music Genre VGG16 (MG-VGG16). The experimental results demonstrated that the MG-CNN model achieved an accuracy of 89.48%, while the MG-VGG16 model achieved an accuracy of 78.93%. Compared to the state-of-the-art methods, the proposed method can significantly improve and facilitate music genre classification tasks.