New target-signal-detection schemes for multi-microphone noise-reduction systems for hearing aids

Poor speech intelligibility in noise is a major source of dissatisfaction for users of both conventional hearing aids and cochlear implants. Many noise reduction schemes have been proposed so far. One of the more promising approaches, the adaptive beamformer, combines the input of two microphones to enhance target signals which are emitted in front of the user, while suppressing noise from other directions. In order to attain satisfactory performance, the adaptive beamformer must be combined with a reliable target-signal-detection scheme to control adaptation. In this work, two new target-signal-detection schemes are proposed and compared with a previously published algorithm by Greenberg et al. [1992]. The proposed delta-sigma algorithm combines a more reliable SNR-estimation with a very low computational load. The new multi-correlation algorithm is computationally more expensive, but enables the user to define the angle of incidence, which differentiates between target signals and noise. To allow further evaluation in everyday listening situations, a real-time version of the adaptive beamformer with the proposed target-signal-detection schemes is being implemented on a portable system