A new pitch tracking algorithm for the spectral modeling synthesis

Abstract

The modeling methods for generating musical sounds are largely categorized into two major sections: physical modeling and spectrum modeling. Spectrum modeling is to parameterize a sound recognizing at the human auditory cognitive mechanism in frequency domain. This paper describes a new pitch tracking algorithm for the spectral modeling synthesis (SMS) technique for Korean traditional percussion instrument, Jing, showing that the pitch tracking technique in conventional SMS cannot represent characteristics of percussion instruments. In contrast, the proposed can guarantee accurate sinusoids and well-approximated residual signals. The peak-picking algorithm improved the conventional SMS by extracting accurate magnitudes of peaks changing irregularly for each frame. The pitch-tracking algorithm enables a pitch continuation to be performed successfully, as well as, softening phenomena and beat phenomena can be interpreted. Compared to the conventional algorithm, the proposed algorithm reduced sinusoidal components in the residual signal significantly, so not only it can synthesize successfully Jing sounds but be also applied to analyze general percussion instruments.