Signal estimators and noise reduction schemes for high-end binaural hearing aids

Abstract

There is a variety of hearing aid models available in the marketplace, which vary in terms of size, shape and effectiveness. Due to size constraints, for certain types of hearing aids only a single microphone per hearing aid can be fitted, and only single-channel monaural noise reduction schemes can be integrated in them. In the near future, binaural hearing aids will become available, making use of a wireless link to shape information between the two ears. In this thesis, a binaural diffuse noise power spectral density estimator is first developed. The estimator does not require a voice activity detection algorithm, it provides an estimate at any time i.e. during speech activity or not, it has no noise tracking latency and it does not assume that the target speaker is in front of the binaural hearing aid user. An instantaneous binaural target speech power spectral density estimator is then developed, for conditions of speech corrupted by lateral interfering noise. The estimator does not require the knowledge of the direction of the noise source, and the noise source can be highly non-stationary or transient. Finally, a complete binaural noise reduction scheme is designed, by incorporating the two binaural estimators mentioned above. The proposed reduction scheme is designed for complex real-life environments composed of time-varying diffuse noise, multiple non-stationary directional noises and reverberant conditions. The proposed scheme also preserves the original interaural cues of both the target speech and the background noises. Simulations using recorded signals provided by a hearing aid manufacturer are performed in the thesis to validate the performance of the proposed estimators and the proposed noise reduction algorithm. The results indicate that the proposed binaural noise reduction scheme proves to be a good candidate for the noise reduction stage of upcoming binaural hearing aids.