Robust acoustic beamforming in the presence of channel propagation uncertainties

Beamforming is a popular multichannel signal processing technique used in conjunction with microphone arrays to spatially filter a sound field. Conventional optimal beamformers assume that the propagation channels between each source and microphone pair are a deterministic function of the source and microphone geometry. However in real acoustic environments, there are several mechanisms that give rise to unpredictable variations in the phase and amplitudes of the propagation channels. In the presence of these uncertainties the performance of beamformers degrade. Robust beamformers are designed to reduce this performance degradation. However, robust beamformers rely on tuning parameters that are not closely related to the array geometry. By modeling the uncertainty in the acoustic channels explicitly we can derive more accurate expressions for the source-microphone channel variability. As such we are able to derive beamformers that are well suited to the application of acoustics in realistic environments. Through experiments we validate the acoustic channel models and through simulations we show the performance gains of the associated robust beamformer. Furthermore, by modeling the speech short time Fourier transform coefficients we are able to design a beamformer framework in the power domain. By utilising spectral subtraction we are able to see performance benefits over ideal conventional beamformers. Including the channel uncertainties models into the weights design improves robustness.Open Acces