Music Genre Classification Using CNN and MobileNetV2

This paper represents a significant advancement in the field of music genre classification and recommendation by harnessing the power of MobileNet’s transfer learning capabilities and custom-designed CNN architectures. By leveraging a diverse dataset and transforming audio samples into spectrograms, the system achieves robust feature representation for input. Through extensive training and fine-tuning processes, both MobileNet and the specialized CNN model are optimized to accurately classify music genres. Evaluation metrics such as accuracy, loss, and confusion matrices provide insights into the performance of the trained models. Upon deployment, the system seamlessly integrates into digital music platforms. Overall, the results demonstrate the efficacy of transfer learning and CNNs in improving genre classification accuracy and enhancing the user experience in discovering and enjoying music. We have achieved remarkable success in music genre classification, with the CNN model attaining an accuracy of $\mathbf{9 0. 2 7 \%}$ and the MobileNet model reaching an impressive $\mathbf{9 8 \%}$.