Wang J., Chang L., Li M.

Abstract The musical instrument, saxophone, is a very important product in Taiwan. In order to minimize the subjective bias during the quality control, an automatic computer system was developed in this work to classify the saxophone sound. From the Fourier analysis using the steady state signal, three parameter can be obtained, i.e., the frequency ratio Fi, amplitude ratio Ai and phase difference Pi of the i-th harmonic partial with respect to the fundamental. The growing rates £_i of the fundamental ant its harmonic partials of each note during the attack were exacted by using the short time Fourier transform. It was found in this work that the parameters Ai, Pi and the £_i were the most important features of timbres of the saxophone. The subjective testing with twenty students indicates that the variation of the parameters (i.e., Ai, Pi, £_i) should be larger than a certain threshold value in order to result in a subjective sensation of timbre change. A parameter called timbre variation TV was defined in this work as: TV= sum (£GVj/ Vtj), j = 1,2¡K,m where m is the number of parameters used for evaluation, and£GVj represents the difference of the parameter Vj between two saxophones which should be compared. The Vtj represents the threshold of the parameter Vj. The subjective testing indicates that the objective value TV is highly correlated with the subjective sensation of timbre difference. Experimental results demonstrate that the saxophone sound can be successfully classified by the single parameter TV proposed in this work. The automatic system was used successfully in (1) quality control when the data base of a standard saxophone was used for comparison, (2) comparison of the timbre difference between the saxophones from two manufactures. Keywords: Saxophone, timbre classification