Application of Deep Learning Model for Music Genre Classification

Abstract

A musical genre is a term used to classify specific musical styles in accordance with a sense of their significance or a set of guidelines. There are many techniques to classify music into different genres, but attempts to automate this process using machine learning have received much less attention from the media. Blues, Classical, Country, Disco, Hiphop, Jazz, Metal, Pop, Reggae and Rock are the most common musical genres. Objective: This study develops and creates a model that can classify music into their respective genre. The dataset, which contains 1000 songs divided into 10 sections with 100 songs each, was obtained from the open-source service "Kaggle" and utilized for the classification. To train and test the model, the dataset was then converted into a JSON format, the new dataset was then splitted into 30% for testing the model and 70% for the training of the model. This model was created using the Convolutional Neural Network (CNN) machine learning technique. The result shows that the model was able to correctly classify 7 audios out of 10 audios and correctly classify them into their original genre. The classifier achieved an accuracy score of about 73%. In conclusion, the study provided a model that will aid in the classification of various types of music into their appropriate genres. This will be highly useful for record companies and other music retailers to group songs for search and retrieval.