Classification of music instruments using wavelet-based time-scale features

Separation of sounds from different sources plays a significant role in success of auditory scene analysis and multimedia content recognition. In this paper, we propose wavelet-based features for discrimination of signals from various music instruments. One hundred and fifty-two music segments from thirteen different instruments were selected from a public music database (Universitat Pompeu Fabra). We performed automatic instrument classification of segments from 13 instruments using selected wavelet features which resulted in accuracy as high as 85%. The wavelet features, along with the considerations suggested and elaborated on here, while are successful for solving the problem at hand, could be applied to many signal processing problems in other domains.