Blind separation of acoustic sources in reverberant environments

In many real-world scenarios, different source signals are mixed in a dispersive environment, and there is significant interest to separate the sources solely on the basis of the observed mixtures, without prior knowledge of the sources or the mixing system. A well-known example is a cocktail party situation, where many people are speaking at the same time in a reverberant room, and a listener has to separate out the signal of interest. Other applications of blind source separation include the ecg analysis of mother and fetus, fMRI data analysis, image restoration, and digital communications over multipath channels. In this project, we aim at finding technical solutions for the separation of sounds in acoustic reverberant environments solely on the basis of observed microphone signals. The project has been funded by the German Research Foundation (DFG).

Project Members

Radoslaw Mazur, Tiemin Mei, Alfred Mertins

A CUDA Implementation of Independent Component Analysis in the Time-Frequency Domain: