Transformer-based Technique to Classify Raags in Hindustani Classical Music

Abstract

The classification of Raags in Hindustani music is challenging because of the non-stationary nature of the music signals. The present research work addresses this issue of the Raags classification. We propose a novel technique based on transformer architecture to classify the Raags from Hindustani Classical music. We have generated a dataset of Hindustani Classical music to evaluate the proposed method. This dataset comprises seven raags of Classical Hindustani Music. We preprocess the acquired data using various prepossessing techniques and the MFCC values are obtained. After that, we use transformer-based architecture to get the information of raags from processed data. As this is the classification algorithm, we use the transformer encoder of the transformer architecture. The performance parameters like precision, recall, and F1 score is used to evaluate the developed technique, which are 0.96, 0.95, and 0.95, respectively. It indicates the superiority of the proposed work over previously published methods in the literature.