Granular Synthesis of Sound Textures using Statistical Learning

We present a statistical learning algorithm for synthesizing random sound textures resembling an input sound texture segment. Our approach begins by constructing a hierarchical multi-resolution representation of the input signal. The resulting tree data structure is then statistically sampled to generate a new tree from which the output sound texture is reconstructed. This method works for both periodic and stochastic sounds and for mixtures of both, without assuming any explicit model for the data. Our results indicate that the proposed technique is effective and robust.