Raga identification using Locality Sensitive Hashing
G Padmasundari, Hema A. Murthy · 2017
Ra̅ga is a quintessential component of Indian classical music. Ra̅gas are primarily characterised by melodic time-frequency (T-F) motifs. There have been several efforts made to determine the identity of a ra̅ga, yet the techniques work only on subset(s) of ra̅gas, or perform poorly in terms of scalability. In this paper, we propose a ra̅ga identification method for Carnatic music using Locality Sensitive Hashing (LSH). LSH is a technique that can match similar time series into the same bucket. LSH is close to how humans perceive ra̅gas, in that it hashes T-F motifs to a smaller subset of ra̅gas. Experiments were performed on a standard dataset consisting of 100 concerts, each with 8-10 ra̅gas, leading to 927 items and 182 ra̅gas. 3000 ra̅ga queries on this database achieved an accuracy of 70%. Most importantly LSH does not suffer from scalability issues of earlier approaches.