Jazz music sub-genre classification using deep learning
Rene Josiah M. Quinto, Rowel Atienza, Nestor Michael C. Tiglao · 2017
Music genre classification is a well-known problem in the field of music information retrieval, with existing machine learning and deep learning solutions [1] [2] [3]. However, solutions for sub-genre classification for a specific music genre are few. This paper shows the boost in performance that Deep Learning techniques can provide in comparison to Machine Learning techniques for sub-genre classification. On a dataset of three (3) sub-genres of jazz music preprocessed using their Mel-Frequency Cepstral Coefficients [4] (MFCC), the most prominent machine learning techniques for genre classification, Neural Networks, SVM, and KNN, achieved a maximum accuracy of 79.39%, 81.67%, and 77.43% respectively, while a single-layered Long Short-Term Memory (LSTM) network achieved an accuracy of 80.30%, and adding a multi-layer perceptron network before the LSTM layer boosted the accuracy to 89.824%.