Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of GraphSearch.
The invention pertains to the field of methods for estimating source signals from a multichannel mixture signal, and the aim of the invention is to carry out a reliable estimation of the properties of source signals without necessarily requiring prior information from the statistic model of each source. Accordingly, the invention relates to a method for separating source signals from a mixture signal (X) corresponding to a combination of source signals (10), said method comprising: a step of acquiring a mixture signal (X), a step of estimating the characterisations of the source signals (z) from the mixture signal (X) and a step of learning source models (Ë) from the characterisations of the source signals (z). The characterisations of the source signals (z) mainly comprise an intermediate estimation of the source signals (s) enriched by auxiliary information (Ë) indicative of the local statistical properties of said intermediate estimation, and the source models (I) describe the general statistical properties of the source signals (10) emitted by the sources present in the mixture. The method further comprises the step of estimating the source signals (S) from the mixture signal (X), the characterisations of the source signals (z) and the source models (Ë).