Development and Objective Perceptual Quality Assessment of Monaural and Binaural Noise Reduction Schemes for Hearing Aids

People with hearing impairment have great difficulties communicating in noisy and reverberant environments. They usually require a higher signal-to-noise ratio (SNR) to achieve the same listening performance as normal hearing people. But even for normal hearing people a noise reduction is desirable in these environments, especially when using modern communication systems such as mobile phones, handsfree devices, or teleconferencing systems. Hence, the development and evaluation of noise reduction algorithms is an active field of research. To achieve the maximum performance and subjective benefit, these algorithms generally need to be evaluated with objective measures that should be based on auditory models in order to predict human perception as closely as possible. The current dissertation contributes to this field by adding yet another dimension to the problem binaural hearing (i.e., listening with two ears). As a starting point of this thesis, some frequently used objective performance measures and novel objective measures based on recent knowledge of the auditory system are reviewed. Using subjective listening tests on signals processed by monaural noise reduction schemes, those measures are identified that exhibit the highest correlation with subjective data. Using these measures, it is possible to optimize single-channel noise reduction algorithms on a perceptual scale. However, the performance of single-channel noise reduction systems is limited and leads to a trade-off between signal distortion and noise reduction. Therefore, multi-channel beamformer algorithms are investigated throughout the main part of the thesis. In general, they introduce less signal distortions and achieve higher noise reduction for non-stationary signals. Typically, beamformers do not have binaural outputs. To provide the user with information on the spatial arrangement of signals, different strategies to partially preserve or reconstruct the binaural information are developed and evaluated with objective and subjective assessment data. For the objective evaluation of the benefit on binaural speech perception, a novel binaural speech intelligibility measure is used and compared to subjective data. For the binaural beamformer schemes, the influence of head diffraction on the performance is analyzed. It is shown, that at least simple head-models should be integrated into the algorithm design to increase the performance compared to traditional free-field designs and to make use of the acoustic decoupling due to the head shadow effects. In order to utilize the algorithm