Selection of the Desired Source for Acoustic Signal Extraction

Proposal for a Bachelor Thesis / Research Internship


Selection of the Desired Source for Acoustic Signal Extraction


In many daily-life scenarios, the Signal of Interest (SOI), e.g., a mobile phone user, a person controlling a smart home assistant or the conversational partner of a hearing aid user can typically only be observed with interfering sources and noise. To estimate the desired source from an observed mixture of the SOI with interfering sources and noise, blind signal extraction techniques, e.g., based on Independent Vector Analysis (IVA), showed good performance at a low computational cost. Without any prior information about the SOI, the blind source extraction algorithm will arbitrarily pick one of the sources from the mixture which may not be the desired one. Hence, the selection of the SOI by incorporating prior knowledge is a crucial component of such extraction methods.

The aim of this thesis is the implementation and evaluation of approaches for SOI selection for signal extraction algorithms based on IVA, where especially the information provided by the time-domain signal envelope should be investigated in more detail ([1] may serve as a starting point). The resulting algorithms should be evaluated and compared in comprehensive experiments for various acoustic environments. As prerequisites, the student should have MATLAB programming experience and affinity to math.



Prof. Dr.-Ing. Walter Kellermann


M.Sc. Andreas Brendel, room 05.018,