Towards a better balance in sinusoidal plus stochastic representation

Sound can be separated into a sinusoidal and a stochastic part, which is useful for transformation and compression. In such a sinusoidal plus stochastic representation, one of the important issues is to obtain a good separation. The paper presents a method to obtain this deterministic-stochastic balance. The method uses an iterative sinusoidal extraction method and a stopping criterion, which indicates when the (important) tonal components are extracted from the signal. Two measures are investigated, with and without psychoacoustic information, to be used in different applications. First, including psychoacoustic information, the stopping criterion is used in the sinusoidal analysis to construct a sinusoidal part that only contains psychoacoustically relevant tonal components. Second, deliberately not including psychoacoustic information, the stopping criterion is used, in a preprocessing stage before the noise analysis, to remove tonal components from the residual. An important advantage of this method is its simplicity and results are promising.