An exploration of genetic algorithms for efficient musical instrument identification

This study explores the use of genetic algorithms (GA) in optimising feature selection for musical instrument recognition. 95 timbral features were used to classify 3006 musical instrument samples into 5 instrument groups. A GA was used to optimise the best selection of features to use with an multi-layered perceptron (MLP) to classify the instruments. Of all the features examined, the Centroid Evolution was found to be the most important. The system was run a number of times with varying numbers of features as determined by the GA. The accuracy of the classi er was not reduced with a reduction in features, indicating that the GA successfully determined the best features to use.