A5-ESR: Multi-channel speech enhancement in highly dynamic acoustic environments

Most multi-channel speech enhancement algorithms achieve significant improvements in static environments but severely degrade in time-variant acoustic scenarios. However, variations are an intrinsic part of the hearing aid application since the hearing device moves with the person who wears it. Especially those algorithms that require the tracking of a large number of parameters are highly susceptible to rapid head movements and other changes in the environment. Furthermore, in scenarios with several target sources, like a conversation at a restaurant table, algorithms which are designed for a single target source lead to significant distortions when several target sources are active. In this project we aim to extend the prediction-driven approach of towards the spatial domain.

Host institution: Ruhr-Universität Bochum (RUB)
Supervisor names: Rainer Martin and Gerald Enzner
Start date: 1.3.2013
PhD enrolment: Yes
Employment contract: Full-time early-stage research fellow, 36 months