Raga Identification Using Convolutional Neural Network
Ankit Anand · 2019 Second International Conference on Advanced Computational and Communication Paradigms (ICACCP) · 2019
Raga is the essence of Indian Classical Music that is used in its composition, performance, improvisation and organization. Automatic identification of the underlying raga in an Indian Classical song has applications in areas like music recommendation and indexing. Unlike the previous approaches to this problem which rely heavily on feature engineering, in this paper, we attempt to develop a Convolutional Neural Network (CNN) that can learn the distinguishing characteristics of a raga from the predominant pitch values of a song. CNNs are a class of deep, feedforward artificial neural networks, usually applied to analyze visual imagery. Experiments were performed on standard datasets of Carnatic music consisting of five and eleven ragas, where the model achieved an accuracy of 96.7% and 85.6% respectively. We also tested the model against allied ragas.