Clustering of ragas based on jump sequence for automatic raga identification

Abstract

Raga is a soul of Indian Music and the Indian Music has been studied on the basis of Raga. It can also be the main parameter for content based retrieval of audio in Music Information Retrieval(MIR) system. In Western music, melody has not many variations, but in Indian music melody has many variations in the form of ragas. Difficulty in understanding the complex structure of raga makes automatic Raga Identification problem a critical one. Many different attempts have been made to address the raga identification problem but realization of such a system for a large collection of raga is a long way to go. There are many issues in regard to this that need to be solved. We have designed solution to tackle two issues in our work and present the methods and the experimental analysis in this paper. The first issue is to handle the large number of raga classes which is solved through the clustering technique based on jump sequence. The second issue that is addressed here is avoiding the calculation of the scale (tonic) of the song to identify the notes in the song. So approaching a method where raga identification can be done even without knowing tonic (scale) of the song. We present the importance of jump sequence in raga identification in this paper. We also present the experimental results on naturally clustering the ragas along with cluster identification for a given song. © 2012 Springer-Verlag.