A novel application of group delay function for identifying tonic in Carnatic music

In this work, we propose a novel application of the group delay function - to identify the tonic pitch in Carnatic music (one of the main sub genres of Indian classical music). Tonic pitch is the reference note chosen by a performer in Carnatic music. Automatic identification of the tonic pitch is essential in Carnatic music in order to perform data driven computational melodic analysis. Group delay functions, which have found numerous applications in speech processing are employed in this task to process pitch histograms to emphasize certain characteristics of Carnatic music that manifest in a pitch histogram, to aid in accurate tonic identification. In order to do so, the pitch histograms are first characterized as the squared magnitude response of a set of resonators in parallel that do not have constant Q. Some interesting properties of the group delay function of this magnitude response has then been illustrated and exploited to identify the tonic pitch. The proposed method is then tested by identifying tonic on a large varied database with an accuracy of 90.70%.