Methods for acoustic data synthesis

Abstract

The paper deals with the problem of synthesizing a given class of sounds from recorded original data. An analysis/synthesis procedure is used, but variation is deliberately introduced in the synthesis step to broaden the set of sounds that are produced. Two methods are described: one that treats the sound as a deterministic signal and one that treats it as a realization of a nonstationary stochastic process. The two methods can be applied separately or used together in a deterministic/stochastic decomposition of the data.<>