A Hybrid Model for Music Genre Classification Using LSTM and SVM

Abstract

With today's cutting edge technology and intractable access to voluminous data files via the internet, it is important to meet the computational needs of every user. Machine learning is one such growing branch of artificial intelligence that has made such demands of the users viable. Machine learning models are paving the way for classification techniques such as in music genre classification, and have shown to be efficient in predicting classes to a great extent. To exploit the time dependent nature of the dataset Long Short-Term Memory (LSTM) Neural Network is used for music genre classification and combined with Support Vector Machine (SVM) classifier to enhance its performance. The hybrid model of these two classifiers resulted into an increase in the accuracy of prediction of the individual models. This hybrid model is imposed on GTZAN music dataset and is compared with the results of standalone models of LSTM and SVM. The proposed model exceeded the independent accuracies of the LSTM and SVM classifiers with an accuracy of 89%, reaffirming the efficient utilization of each classifier.