Pengembangan Aplikasi Klasifikasi Sepuluh Genre Musik

Abstract

This study aims to classify music by genre, which is used to group music by genre to make it easy to search. The data used audio data and then converted into melspectogram images of 10,000 data divided into ten genres. They are blues, classical, country, disco, hiphop, jazz, metal, pop, reggae dan rock. The data validation using the split test with the training and testing data ratio is 9:1. The method used in this study is a Convolution Neural Network (CNN) using the Tensorflow Keras library. This study conducted experiments with three different architectural models to compare and find the best model. Based on the results, the model that produced the best test accuracy value was the CNN_2 model, with an accuracy value of 81.7% in the test data and the accuracy of the training data of 95.5% by undergoing an epoch of 120 and with a configuration using adam optimizer, loss categorical-cross entropy and learning rate of 0.00005.