North Indian Classical Instrumental Raga Music: Multifractal Parameters Analysis and Raga Recognition System

Abstract

Introduction: The foundation of Indian classical music is the raga. Ancient music literature describes raga as a collection of particular arrangement of musical notes, that turns into soothing music. According to basic music theory, the raga segments have fractal properties. In fractal theory, the scale of the music note (signal) can be altered while maintaining its shape
Objectives: This enhances musical quality with improved raga note. In the proposed work, Codebook of Feature (CoF) model is used to recognize 12 Indian classical ragas. The application considered is north Indian classical instrumental raga music. The multifractal parameter analysis of musical segments is done, based on fractal theory.
Methods: The proposed work recognizes raga without note detection, hence reducing the complexity of the raga recognition system. The training and testing datasets considered are 60 % – 40 % and 90 % – 10 % respectively.
Results: The accuracy obtained is 98.94 % and 99.01 % respectively. The accuracy calculated based on F1 – score for the mentioned datasets is 98.93 % and 99.06 % respectively.
Conclusions: The proposed system is also compared with the recognition of raga using previous work on the same dataset that was implemented with variants of Mel Frequency Cepstral Coefficients (MFCC) features and the ensemble bagged tree as a classifier, which gives 96.32% accuracy. The proposed system's accuracy has increased by 3% compared to MFCC features.