Adding Dynamic Smoothing to Mixture Mosaicing Synthesis
Graham Coleman, Jordi Bonada, Esteban Maestre · 2011
Abstract—Recent works in sound mosaicing synthesis [1], [2] have proposed algorithms that permit instantaneous mixtures of several sources atoms, based on sparse signal representation techniques. We propose combining l1 regularization with linear dynamical smoothing as in the Kalman filter (also in [3], [4]) to promote desired transitions between atoms, while adapting the generic approach to the mixture mosaicing context. Furthermore, we modify the dynamics cost slightly to further promote sparse scores in the case of non-negativity. This is a work in progress in which we can present some sound examples, but for which the proposal is not fully validated. I.