Blind Source Separation

In many situations in daily life, we are exposed to a medley of signals (various speakers, music in the background, noise, etc.). Yet, we are capable of focussing on the particular source we want to hear despite having no a priori information about these signals. Blind Source Separation is a research area that attempts to mimic this human capability using microphone arrays for sound capture and signal processing algorithms for separating mixtures into their individual components. The applications of this technology are wide ranging: hands-free speech systems, hearing aids, robust speaker and speech recognition, etc., to name a few.




The Coctail Party Problem


Projects/theses in this field shall focus on the study of the underlying nature of the problem, leading to the development of new algorithms and/or the implementation of existing algorithms for source separation. Depending upon the topic, the algorithms could be implemented for offline use or in real-time.


3-D Beam Pattern of a Microphone Array

Prerequisite for these topics is a basic knowledge of signal processing. A basic understanding of statistics and linear algebra and programming skills in MATLAB and/or C++ are advantageous, but not compulsory.

Contact: Alexander Schasse