Binaural beamforming taking into account spatial release from masking

Hearing impairment is a prevalent problem that comes with many daily life challenges. These challenges, mainly in the areas of speech intelligibility and sound localisation, can have consequences ranging from social isolation to physical danger. Even though current hearing aid technology can partly alleviate these issues, in practice many acoustic scenes are still challenging. One of the shortcomings of spatial filtering in hearing aids is that speech intelligibility is often not optimised for directly, as that is a subjective measure and more difficult to quantify. Instead, most developed beamformers focus on maximising the speech-to-noise-plus-interference ratio (SNIR), which is only one factor that influences intelligibility. The psychoacoustic effect known as spatial release from masking (SRM) is usually not considered, but can also be a dominant factor.

In this paper, a signal model is developed that explicitly takes SRM into account in the beamforming design. This is achieved by transforming the binaural intelligibility prediction model developed by Beutelmann and Brand (2006, JASA 120:331) to a signal processing framework. The main phase of this model, the equalisation–cancellation (EC) phase, can be represented as an internal beamformer that accounts for the spatial filtering of the auditory system. Internal masking noise is added to the hearing aid signals to model hearing thresholds, which is used to personalise the model. When concatenated with a typical beamformer signal model, the output of this extended model can be used to analyse existing beamformers and design new beamformers one step closer to how the auditory system perceives binaural sound.

It can mathematically be shown that the binaural minimum variance distortionless response (BMVDR) beamformer, which is known to be optimal in maximising the SNIR of the beamformer signals, is also an optimal solution for the extended, perceived model. This seems to suggest that SRM does not play a significant role in improving intelligibility after optimal beamforming is already performed. What is different compared to the classic case, however, is that the solution is no longer unique; the solution space has a number of degrees of freedom dependent on the number of microphones in the hearing aids. These degrees of freedom can be used to preserve binaural cues of interferent sources, while still achieving the same perceived performance of the BMVDR beamformer. The proposed beamformer might in practice be sensitive to intelligibility model mismatch errors, and the practical performance needs to be studied in more detail.