Musical Instrument Classification Based on Improved Matching Pursuit with Instrument-Specific Atoms

Abstract

Taking advantage of the features that atomic parameters describe the essential characteristics of musical instrument, we propose an improved matching pursuit (MP) algorithm with instrument-specific atoms for musical instrument classification. Firstly, through experiment and mechanism analysis, it is observed that the parameters of atoms from sparse decomposition are closely related and have close ties with the structure and vocal mechanism of each musical instrument. Secondly, an improved matching pursuit algorithm is proposed by analyzing the regularity of the atomic parameters of each specific instrument, which simplifies the extraction process of the scale parameters and narrows the range of the frequency parameters of Gabor atoms in traditional MP algorithm. Finally, the atomic parameters extracted from the improved MP algorithm are fed to the support vector machines (SVMs) for identification and classification. Simulation experimental results show that the time cost of the proposed algorithm is about 1/3 of the traditional one, while the recognition rate only decreases from 89.17% to 87.44%.