Identification of Indian Musical Instruments by Feature Analysis with Different Classifiers

Abstract

Most of the musical instrument identification systems that are now available are operated on isolated notes of western musical instruments. But significant work is not yet done on Indian musical instruments. There is a need of extensive study of Indian musical instruments for various applications in different areas.This work presents a system for identifying a specific musical instrument from monophonic recordings. The system proposed in this paper has been trained and tested with three Indian musical instrument samples. Instruments include flute, harmonium and sitar, which are most commonly used in Indian classical music. This work is organized in two stages. The first stage explores extraction of the feature set from musical instrument samples. The feature set includes statistical, temporal and spectral, cepstral, and linear predictive features. In second stage, for identification purpose system has been experimented with k-Nearest Neighbor (KNN) classifier and Feedforward backpropagation neural Network (FFBNN) and the comparative performances were recorded. The system has shown average identification accuracy of 91.67% for combination of linear predictive coefficients (LPC) and K-nearest neighbor (KNN) classifier. For the combination of LPC and Feedforward backpropagation neural network, system has shown 99.37% accuracy.