Generalized Multichannel Wiener Filter for Spatially Distributed Microphones

Abstract:
Considerable noise reduction can be achieved with the multichannel Wiener filter (MWF), which aims to estimate an unknown desired signal. In practice, the standardMWF (S-MWF) estimates the speech component in one of the microphone signals, referred to as the reference microphone signal. Recently, a different formulation of theMWF, which uses the envelope of all acoustic transfer functions (ATFs) between the speech source and the microphones, has been presented. It has been shown that this formulation of the MWF leads to a higher broadband output SNR than the S-MWF, especially for spatially distributed microphones. In this paper, we show that all ATFs can be exploited in a more general way to derive a generalized formulation for the MWF (G-MWF). This generalized formulation can then be used to derive an alternative expression for the S-MWF, where experimental results show that this alternative formulation leads to similar results as the G-MWF using the envelope of all ATFs.