Beamformer for Driving Binaural Speech Enhancement

A beamformer for binaural speech enhancement systems in digital hearing aids is proposed. Its single modules for the estimation of the time-difference-of-arrival (TDOA) and time-alignment operate in the frequency-domain and have a low computational complexity. The TDOA estimation is performed efficiently by a generalized cross-correlation with phase transform weighting. The estimation accuracy for filter-banks with a limited number of subbands, which are needed for hearing aids to meet tight delay constraints, is improved by a histogram-based TDOA estimation. The subsequent time-alignment is accomplished by a simple multiplication with spectral phase factors. A primary application of the proposed system are binaural cue preserving speech enhancement systems based on spectral weighting. The proposed beamformer can be used as delay-and-sum and/or delay-and-subtract beamformer to provide subband signals from which the power spectral densities of interfering sources can be estimated to drive the spectral weight calculation.