A matrix-based approach for evaluation of vocal renditions in hindustani classical music

Abstract

Indian or Hindustani Classical Music (ICM and HCM, respectively) is based on the Raga system of music. Unlike a Western Classical Music (WCM) composition (usually documented in a written score format), a Raga defines an overall melodic structure to which a vocal or instrumental rendition adheres to. HCM has around 200 Ragas, each of which is specified in terms of a subset of notes (that can be part of a composition in that particular Raga) as well as their sequencing (e.g. aroha–avaroha sequence), combinations and permutations. The rules and conventions (here under referred to syntax of Raga) together determine the general melodic structure (often referred to as a mood that the Raga aims to capture) of a rendition. Hence, a computational model for recognition or evaluation of HCM renditions would need to be able to handle a much higher degree of complexity than a similar system for WCM. Here we present a computational model that handles this complexity elegantly and efficiently for 8 Ragas (that are typically learned by novice singers). More specifically we show how a relatively simple note transition matrix-based approach incorporating key elements of a Raga’s syntax results in highly reliable evaluation of songs and robust error identification (for feedback to novice learners as part of computer-based tutoring system).