Enhancing Music Genres Classification with MFCC and CNN

Abstract

Music genre classification aims to group music genres with a high degree of accuracy. Music genre classification is a critical challenge in pattern recognition and digital signal processing. In this paper, we introduce music genres classification using Mel-Frequency Cepstral Coefficient (MFCC) as an extraction feature and using the algorithm Convolutional Neural Network (CNN) as a classification model. The MFCC feature was chosen because of its ability to represent the frequency characteristics of audio signals that correspond to human auditory perception, where the music genre dataset was processed into an MFCC representation before being trained on a CNN model. In this study, we compare three different CNN model to determine the best architecture. The results showed that model architecture 1 obtained the best accuracy during training at 97.15%, while model architecture 2 obtained a training accuracy of 95.74% and model architecture 3 obtained a training accuracy of 95.18%. In testing with new data, model architecture 3 obtained the highest accuracy compared to the other 2 models, with 81%, which indicates good generalization ability. This study proves that the combination of MFCC and CNN is effective for music genre classification with high accuracy.