Extended Source-Filter Model of Quasi-Harmonic Instruments for Sound Synthesis, Transformation and Interpolation

In this paper we present a new technique for sample-based sound synthesis. The approach comprises the analysis of sounds of an instruments sound database, a parameter estimation for an instrument model and a sound synthesis using this model together with the analyzed sounds. The analysis of the sounds is carried out by the segregation of each sound into a sinusoidal and noise component and extracting certain control parameters from both. The components will be modeled using an extended source-filter model, whereas the harmonic component will be represented by a non-white source and a resonator filter and the noise component by a single filter. Model parameters are represented by means of weights of tensor product B-splines (basic-splines) covering the instruments sound characteristics over its full pitch range, global intensities and the sounds temporal evolution. This structured sound representation will allow enhanced source filter based sound manipulations. The paper concludes with a subjective evaluation presented for comparison with state of the art sound transformations.