Musical instrument classification using higher order spectra

This paper presents classification and recognition of monophonic isolated musical instrument sounds using higher order spectra such as Bispectrum and Trispectrum. Experimental results on a widely used dataset shows that higher order spectra based features improve the recognition accuracy, when combined with conventional features such as Mel Frequency Cepstral Coefficient (MFCC), Cepstral, Spectral and Temporal features. Nineteen western musical instruments covering four families with full pitch range have been used for experimentation.

1. Introduction

The use of digital music on internet is increasing day by day. It would be difficult to choose and classify which music and sound to listen from large database. It would be a great help to musicians to provide Music search engine which classify their music sound to listen. In this paper, Musical instrument classification and recognition has been implemented using Higher Order Spectra and other conventional features using Self Organizing Map supervised neural network.