Sound hybridization techniques based on a deterministic plus stochastic decomposition model

Abstract

Current sound hybridization techniques are ha.sed on the exciter-resonator model, where the spectral characteristics of a sound (resonator) are applied to another one (exciter). The tradi­ tional implementations use Linear Predictive Coding, the Short-time Fourier Transform (STFT) or the Phase Vocoder and the most common application is to hibridize speech (used as a res­ onator) with a spectrally dch sound (used as a exciter), resulting into the characteristic "talking sound" effect. The use of a more powerful sound representation model like Spectral Modeling Synthesis (SMS)[Serra and Smith, 1989] takes the concept of sound hybridization beyond the traditional exciter-resonator model. This paper starts by describing a hybridization system ha.sed on the STFT and then it presents the system ha.sed on SMS.