Abstract

The use of multichannel speech enhancement systems offers some significant advantages over single channel solutions. Speakers can be localised, tracked and unwanted noise can be spatially supressed to improve perceived audio quality in recordings or for telephone conference participants. Differential microphone arrays have gained a lot of attention in the last years because of their ability to keep their directional response pattern constant over a wide frequency range and their small sensor spacings.

Focusing on circular geometries this thesis will explore algorithms to localise speakers and electronically steer a beamformer to improve the quality of the speech signal. The first part of the thesis is focused on the design of data independent time-invariant beamformers for circular geometries. Different design algorithms will be shown and their advantages and drawbacks will be discussed.

In the second part the possibilities of sound source localisation with small aperture arrays will be explored.

To test the proposed algorithms for beamforming and localising, real world measurements were done in a reverberant environment. The performance of the proposed algorithms is evaluated and errors of the localisation algorithms and noise suppression ability of the beamformers are measured.