Downstream speech enhancement in a low directivity binaural hearing aid
Marinus M. Boone, Opdam, Robertus Cornelis, Schlesinger, Anton · RWTH Publications (RWTH Aachen) · 2010
The combination of a binaural beamformer with an auditory model-based localizer and post-processor is presented. The intention of the design was to maintain a high degree of quality and listening ease while enhancing speech intelligibility. Therefore, a binaural and fixed minimum variance distortionless response beamformer was employed to establish a medium directivity at low self noise and high binaural naturalness. By applying models of binaural interaction at the two-channel output, sound sources are localized and separated. The process of binaural interaction was extended by an auditory model of across frequency interaction in the localizer and a model of modulation perception in the post-processor. To adapt the post-processor to the scene and to keep the introduced distortion at a low level, the parametric output of the localizer was used to steer the aperture of a spatial filter during post-processing. For that reason, Bayes theorem was applied to calculate the a posteriori probability of the target source in a complex acoustic scene and to use this probability for the formulation of a data-driven filter. The localizer and post-processor were assessed in a range of acoustic arrangements using an objective intelligibility and quality measure for nonlinearly processed speech. The results show a significant improvement in situations with one interferer and no decline of intelligibility and quality in more complex situations.