Automatic Tonic (Shruti) Identification System for Indian Classical Music

Abstract

In Indian classical music, identification of basic pitch is a crucial task to explore information of full raga. Here recorded dataset of both instrumental and vocal raga excerpts is used. Pitch frequency is calculated using subharmonic-to-harmonic ratio. Pitch estimation using SHR is tested for nine different instruments, and it shows 95.06% pitch detection accuracy which is better compared to other methods. Each frame of the raga is processed to increase efficiency in the tonic identification. Pitch histogram is used to extract various features of the raga. The various machine learning algorithms are tested, and the best algorithm is used to find set of rules for the evaluation of Tonic. The very famous J48 decision-tree algorithm shows highest test accuracy of 92.86% for tonic identification. This test model is further used to build an iterative system for tonic identification with the highest confidence. The proposed system is tested for two datasets. Tonic identification accuracy for the first dataset is 90.5%, and it is 93.05% for the second dataset.