A Hierarchical Stratagem for Classification of String Instrument

Associated Research Activities

Canvassers have paid efforts on several aural traits to discriminate instruments. Haralick (1992) has suggested a method for extraction of statistical features in the meadow of processing of images. These statistical features are very useful and they may be utilized in other domains too. Most of these works have used MFCC. A spectral trait for musical gadget categorization has been used by Agostini et al. (2003). Centroid bandwidth, pitch, skewness as well as zero crossing rates have been used by them. Peeters (2003) has proposed his own algorithm named IRMFSP for large database of musical database. Spectral features were also been explored by some researchers like Zhu et al. (2004). They have used spectrum of instruments. Their work was limited to jazz, pop and rock instrumentals only. Kaminskyj and Czaszejko (2005) have been able to classify mono type musical instrumental sounds with the help of 6 traits - cepstral coefficients, constant Q transform frequency spectrum, multidimensional scaling analysis trajectories, RMS amplitude envelope, spectral centroid and vibrato. Algorithms for automatically categorization of musical instrumental sounds have been proposed by Benetos, Kotti and Kotropoulos (2006). Hierarchical scheme was ventured by Essid et al. (2006). Support Vector Machine or SVM has been used there. They have used spectral features like MPEG-7 audio features, cepstral traits for example Mel Frequency Cepstral Coefficients or MFCC, temporal traits like autocorrelation coefficients and ZCR, wavelet traits, perceptual traits for example sharpness and loudness.