Binaural beamforming with spatial cues preservation for hearing aids in real-life complex acoustic environments

This work is introducing novel binaural beamforming algorithms for hearing aids, with a good trade-off between noise reduction and the preservation of the binaural cues for different types of sources (directional interfering talker sources, diffuse-like background noise). In the proposed methods, no knowledge of the interfering talkers' direction or the second order statistics of the noise-only components is required. Different classification decisions are considered in the time- frequency domain based on the power, the power difference, and the complex coherence of different available signals. Simulations are performed using signals recorded from multichannel binaural hearing aids, to validate the performance of the proposed algorithms under different acoustic scenarios and using different microphone configurations. For the simulations performed in this paper, a good knowledge of the target direction and propagation model is assumed. For hearing aids, this assumption is typically more realistic than the assumption of knowing the direction and propagation model of the interferer talkers. The comparison of the performance results is done with other algorithms that don't require information on the directions or statistics of the interfering talker sources and the background noise. The results indicate that the proposed algorithms can either provide nearly the same noise reduction as classical beamformers but with improved noise binaural cues preservation, or they can produce a good trade-off between noise reduction and noise binaural cues preservation.