Let's Identify the Similar and Confusing Raags of Hindustani Classical Music

The identification of different raags in Indian classical music is a challenging task, and distinguishing between Raag Bhoopali and Raag Deshkar is no exception. The reason why this is a difficult task is both the raags share similar melodic structures, including the use of a pentatonic scale (“Audav” scale) with five notes (Sa, Re, Ga, Pa, and Dha). This makes it challenging to differentiate between the two raags, especially for those who are not familiar with the nuances of Indian classical music. The present research work addresses this issue of identification between two raags. We have purposed a method based on attention employing a Neural Network (NN) architecture, for identifying the two. To assess the proposed technique, we curated a dataset of Hindustani Classical music comprising these two distinct raags. Before proceeding with the analysis, we performed multiple pre-processing steps, audio data augmentation techniques, and feature extraction methods on the dataset, which yielded several feature values. Next, we employed the proposed architecture that utilized a classification algorithm. Additionally, to gain further insights into the proposed approach we conducted a counterfactual analysis, which involved identifying the minimum changes required in the input data to change the classification outcome. The attention-based neural network architecture plays a significant role in extracting pertinent information about the raags from the pre-processed data. The developed technique was evaluated based on key performance parameters such as accuracy, precision, recall, and F1 score, yielding impressive scores of 0.949, 0.949, 0.949, and 0.949 respectively. This indicates the clear superiority of our proposed approach over previously published methods in the literature.