Novel audio feature set for monophonie musical instrument classification

This paper proposes a novel set of parameters for musical instrument classification of three different instrument classes of the instrument from audio recordings of monophonie musical sounds notes. The proposed method extracted three features: attack slope, constant Q transform and cepstral coefficients. The algorithm consisted of feature extraction and classifier learning steps. Thus, this system depends upon low-level features. A maximum accuracy of 87% was obtained when tested on the database obtained from University of IOWA Electronic Music Studios. Confusion matrix and subsequent significant classification metrics have been computed. This classification scheme finds application in audio indexing, content based retrieval, genre identification of instrumental music.