Predominant Musical Instrument Classification based on Spectral Features

Abstract

With the aim to examine one of the cornerstone problems of Musical Instrument Retrieval (MIR), a spectral feature-based methodology for the classification of predominant instruments used in an audio sample is presented. For this purpose, the IRMAS dataset has been chosen. It includes clips of 3846 music samples with around 192 minutes run-time recorded from various sources in the last century, spanning multiple genres like country folk, classical, pop-rock, Latin-soul etc., making the data set diverse and better training.