Enhanced automatic source identification of monophonic musical instrument sounds
Ian Kaminskyj, P. Voumard · 2002
A multistage intelligent hybrid classification system is proposed which will automatically identify the source of monophonic sounds for up to 19 different musical instruments. Features to be evaluated for their effectiveness in instrument classification include waveform amplitude envelope, constant Q frequency spectrum, spectral onset asynchrony, spectral onset peak position asynchrony, harmonicity/overtones, brightness and articulation. Principal component analysis will be performed on this feature set as a means of dimensionality reduction. An artificial neural network and a nearest neighbour classifier will be compared to determine which provides optimum classification ability. A rule based expert system will be utilised to combine note identification information with the instrument classification task.