A Novel Approach To Music Genre Classification Through Deep Learning

All kinds of music genres are classified into some common categories such as pop, electronic dance music, metal, jazz, blues, country, rock, and hip-hop. The most demanding task in the music information retrieval (MIR) process is organizing different music files based on these genres. Automatic music genre classification is a very important study for music platforms such as Spotify, Wynk, and Saavn that have a great impact in terms of personalized recommendations and offerings. Today’s deep learning techniques like Convolutional Neural Networks (CNNs) are very successfully applied to a large variety of data by learning and inferring patterns and trends. Different techniques are reviewed for music genre classification using CNN. Further fine-tuning and additional development of VGG16 would help improve classification accuracy by learning complex feature representations of audio spectrograms and being trained on initial datasets. It is indicated that VGG16 is efficient for learning genre-specific patterns and so can serve as a flagship framework for automatic music genre classification. To improve automatic classification of music genres, ResNet-18 with the residual learning architecture was also tried for boosting classification performance in automatic music genre classification. The efficiency with which ResNet-18 can avoid the vanishing gradient problem because of the skip connections further improves the lend of deep architectures and advances automatic classification of music genres.