Dereverberation with differential microphone arrays and the weighted-prediction-error method

Reverberation is one of the most detrimental effects on the quality and performance of hands-free speech communication and human-machine interfaces. Among different approaches investigated in the literature, the weighted-prediction-error (WPE) method has exhibited some promising potential for dealing with reverberation. However, the WPE is highly sensitive to additive noise and the presence of even small levels of noise can cause significant performance degradation to WPE. To improve the robustness of this technique, we study in this paper the combination of WPE and differential microphone arrays (DMAs), resulting a DMA-WPE method. Our approach is as follows. A DMA is used to pick up the sound signal of interest in noisy and reverberant environments. The microphone signals are processed with a differential beamformer to suppress noise, interference, and reverberation. The output of the differential beamformer is then used as the reference signal for WPE to further perform dereverberation. Simulation results demonstrate that the DMA-WPE method outperforms WPE significantly in all the studied conditions.