Tonic Independent Raag Classification in Indian Classical Music

Sathwik Tejaswi Madhusdhan, Girish Chowdhary

Keywords: Deep Learning, Machine Learning, Music Information Retrieval, Spatio-temporal Data

TL;DR: This paper aims at robust pattern recognition in Indian Classical Music

Abstract: A vital aspect of Indian Classical music(ICM) is the Raag, which serves as a base on which improvisations and compositions(IAC) are created and presented. While every Raag represents a unique emotion, it allows a musician to explore and convey his interpretation of the lyrical content of a song. Although many works have explored the problem of classification of Raag, they have several short comings owing to the fact that they assume a prior knowledge of the of the audio or is dependent on a preprocessing technique that identifies the tonic. In this work we introduce a novel data augmentation technique leveraging an inherent aspect of ICM that the semantics of IAC are only dependent on the relative position of notes with respect to the tonic and not the tonic itself Convolutional Neural Network based approach to build a robust model that can classify Raag independent of the tonic.