This paper deals with the separation of two convolutively mixed signals. The proposed approach uses a recurrent structure adapted by generic rules involving arbitrary separating functions. While the basic versions of ...
详细信息
This paper deals with the separation of two convolutively mixed signals. The proposed approach uses a recurrent structure adapted by generic rules involving arbitrary separating functions. While the basic versions of this approach were defined and analyzed in our companion paper (Charkani and Deville, 1999), two extensions are considered here. The first one is intended for possibly colored signals. In addition, the second one may be used even when the probability density functions of the sources are unknown. We first analyze the convergence properties of these extended approaches at the separating state, i.e. we derive their equilibrium and stability conditions and their asymptotic error variance. We then determine the separating functions which minimize this error variance. We also report experimental results obtained in various conditions, ranging from synthetic data to mixtures of speech signals measured in real situations. These results confirm the validity of the proposed approaches and show that they significantly outperform classical source separation methods in the considered conditions. (C) 1999 Elsevier Science B.V. All rights reserved.
In this paper, we investigate the self-adaptive source separation problem for convolutively mixed signals. The proposed approach uses a recurrent structure adapted by a generic rule involving arbitrary separating func...
详细信息
In this paper, we investigate the self-adaptive source separation problem for convolutively mixed signals. The proposed approach uses a recurrent structure adapted by a generic rule involving arbitrary separating functions. A stability analysis of this algorithm is first performed. It especially applies to some classical rules for instantaneous and convolutive mixtures that were proposed in the literature but only partly analysed, The expression of the asymptotic error variance is then determined for strictly causal mixtures. This enables to derive the optimum separating functions that minimize this error variance. They are shown to be only related to the probability density functions of the sources. To perform this error minimization, two normalization procedures that improve the algorithm properties are proposed. Their stability conditions and their asymptotic behaviour are analysed. (C) 1999 Elsevier Science B.V. All rights reserved.
暂无评论