Hindustani and Carnatic music (HM, CM) are India's two predominant art music traditions. This talk discusses the computational analysis of percussion in these two traditions, focusing mainly on solo performances of mridangam and tabla. In CM, a section of a concert is dedicated to percussion instruments called taniavartanam, whereas in HM, separate solo tabla concerts exist.

This talk discusses our work on structural segmentation and labelling of tabla solo and taniavartanam recordings. Segmentation refers to the task of automatically locating the significant changes in the rhythmic structure. We perform this task by analyzing the local self-similarity in the recording in an unsupervised manner. For tabla solos, we also use a supervised random forest and a convolutional neural network trained on hand-crafted features inspired by the distinct properties of tabla solos.

Our study used 35 hours of tabla solo recording and 12 hours of taniavartanam and achieved reasonable accuracy in the segmentation task. This work enables musical understanding and description of taniavartanam and tabla solo concerts.