Binaural Hearing Aid Using Sound-Localization-Preserved MMSE STSA Estimator with ICA-Based Noise Estimation

摘要

In this study, we first propose a method of blind speech extraction for a binaural hearing aid combining ICA-based noise estimation and the MMSE STSA estimator of the target speech. Next, to improve the sound localization accuracy as well as the noise reduction ability, we theoretically derive the equi-binaural optimal spectral gain that minimizes the residual noise power in terms of the MMSE under the constraint that sound localization images are preserved. We conduct experiments to evaluate the performances on noise reduction, sound distortion, and perceptual localization using actual recorded head-related transfer functions and environmental noises. Compared with conventional ICA, the proposed method is superior regarding both noise reduction and sound localization.