Chair of
Multimedia Communications and Signal Processing
Prof. Dr.-Ing. André Kaup

Preservation of spatial cues in blind source separation schemes

Supervisor:Dr.-Ing. Robert Aichner
Faculty:Prof. Dr.-Ing. Walter Kellermann
Student:Meray Zourub
Info:Blind source separation (BSS) addresses the problem to separate sources from a set of linear mixtures. In the acoustical scenario this corresponds to multiple speech signals acquired by a microphone array. Thus, the mixing system is convolutive, i.e., it consists of room impulse responses which can be modeled by finite impulse response (FIR) filters. To accomplish separation we have to assume that the source signals are mutually independent, which is in general true for speech sources. The term ``blind'' stresses the fact that no additional a priori knowledge about the source signals, mixing conditions, or the sensor array configuration is necessary.

One application of BSS techniques is the separation of a desired source from the interfering sources in noisy environments which is also termed the “cocktail party problem”. Therefore, BSS algorithms are desirable, e.g., in binaural hearing aids. In such applications, the BSS algorithms should in addition to high noise reduction also preserve spatial cues which are mainly described by interaural time and level differences between both ears of a listener. However, conventional BSS algorithms yield monaural representations of the separated signals and thus, no spatial characteristics of the signal can be perceived.

In this master thesis a new technique to recover the spatial impression of sound signals from a single monaural BSS output containing the desired signal should be investigated. The algorithm should be implemented in Matlab and should be tested in various acoustic environments using objective measures and by performing informal listening tests. Possible limitations should be identified and modifications which may lead to a more robust algorithm should be investigated. Moreover, relationships to a recently presented BSS approach which yields binaural representations of each separated output signal and thus preserves the interaural time and level differences should be examined. Possible advantages of this BSS approach should be identified and incorporated in the investigated technique.

A clear and comprehensive documentation of both, the software and the experimental results is of utmost importance.

TypeMaster Thesis