Identification of Carnatic raagas using Hidden Markov Models

Abstract

Raaga identification is one of the key areas for budding Carnatic musicians and avid listeners. Identification and knowledge of the raaga of a song not only implies knowledge of music but also helps establish the mood of a song. We propose to identify a Carnatic raaga by extracting from the music sample, information about the 12 distinguishable frequencies in an octave. The proposed technique is Specmurt analysis which involves the analysis of a signal in its log-frequency domain. The extracted information is fed to the Hidden Markov Model back-end system where each raaga has its associated model.