Carnatic Classical Raga Identification Using Stacked Ensemble Learning Model
Arunya Paul, Sasmita Pahadsingh, Tejaswini Kar, Shruti · 2024
This paper presents an innovative approach for Raga Identification in Carnatic Classical Music, leveraging a Stacked Ensemble Learning Model alongside individual classifiers, namely Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF). The dataset encompasses diverse ragas, both Melakartha and Janya, with features like Mel-Frequency Cepstral Coefficient(MFCC), Spectral Centroid, and Chroma Short Time Fourier Transform. The individual classifiers, providing accuracies 96.718% for KNN, 96.878% for MLP, 96.875% for RF, and 97.031% for SVM, demonstrate commendable performance. The Stacked ensemble mode however outperforms individual classifiers, achieving an accuracy of 97.812%. Result analysis incorporates precision, recall, F1 measure, accuracy, and support metrics, illustrating the ensemble's superior performance in capturing the intricate nuances of ragas. This paper not only contributes to the burgeoning field of computational music analysis but also sheds light on the efficacy of ensemble learning for nuanced Raga Identification. The findings underscore the potential of the proposed model to accurately classify diverse ragas, paving the way for advancements in automated recognition systems for Indian Classical Music. Future work may explore hyperparameter tuning and the generalization of the proposed model across varied datasets and musical renditions.