Automatic classification of musical instrument sounds

Abstract

The aim of the present study is to show that the process of automatic classification of musical instrument sounds is possible on the basis of a limited number of parameters. However, due to the complexity as well as to the unrepeatable nature of musical sounds, both steady and transient states should be taken into account while creating feature vectors. For this purpose a database of musical instrument sounds was built containing various instrument sounds played with a different articulation. Then, this database was used in further experiments consisting of some stages, i.e., preprocessing, parametrization, and pattern recognition. The main subject of this study was the optimization of the set of parameters to be included in the feature vectors. Therefore, the quality of parameters with regard to automatic pattern recognition was analyzed both using statistical methods and learning algorithms.