Multichannel Wiener filter estimation using source location knowledge for speech enhancement

In this paper a technique for estimating the single channel Wiener filter post-processor using two complementary adaptive near-field beamformers is presented as an alternative to voice activity detection for speech enhancement applications. Two near-field beamformers, the MVDR beamformer and an adaptive nullformer based on noise to signal maximisation, are used to generate estimates of signal and noise statistics which can be used to compute an estimate of the single channel Wiener filter for noise reduction. It is demonstrated that the performance of the estimated filter compares well with the perfect Wiener filtering case, and shows good improvement in speech intelligibility.