ISBN 978-3-8439-0837-5

978-3-8439-0837-5, Reihe Ingenieurwissenschaften

Stefan Wehr
Robust Binaural Blind Source Separation in Hearing Aids

182 Seiten, Dissertation Universität Erlangen-Nürnberg (2012), Softcover, A5

Zusammenfassung / Abstract

In this thesis, a special kind of spatial filtering is investigated for the application in binaural hearing aids: Blind Source Separation (BSS). In contrast to conventional beamforming techniques, BSS is capable to deal with both unknown microphone positions and unknown source positions in so-called cocktail party situations, where multiple sources are simultaneously active. Moreover, BSS is able to handle arbitrary obstacles in between the microphones. After introducing the considered BSS algorithms, both the sound propagation from the (external) sources to the ears as well as the voice of the hearing-impaired person are investigated. Moreover, the problem of identifying the target source is addressed. By exploiting a BSS-based source localization algorithm, the target source can be automatically selected by the look direction. Furthermore, a new concept for providing fast and robust source tracking is presented. This concept allows for both source separation and source tracking in rapidly time-varying scenarios. Additionally, a wide spectrum of algorithms for binaural reproduction is discussed. This diversity allows for balancing between the computational complexity and the quality of the binaural reproduction. Finally, the investigations are extended to the voice of the hearing-impaired person, where both the source separation performance and the binaural reproduction are considered.