Rough Set Based Automatic Classification of Musical Instrument Sounds

Automatic classification of audio data arose increasing interest recently. This paper addresses the problem of automatic recognition of musical instrument sounds, applying rough set based techniques as a tool of classification. Instruments representing wind and string families were used in the experiments. Since the main problem in case of audio data is the proper parameterization, we also investigated issues regarding various parameterization methods. Fourier transform and wavelet analysis were applied as parameterization tools. The obtained feature vectors were tested using rough set tools. The analyzed data represent singular sounds of full musical range of 11 musical instruments, played with various articulation techniques. Results of experiments are presented and discussed in this paper. We summarize our paper with conclusions on musical signal representation for timbre classification purposes.