A two-stage multichannel acoustic noise reduction under reverberant environments

In this paper, there are two contributions for multichannel noise reduction techniques under reverberant environments. The first contribution is a theoretical analysis of previously proposed multichannel spatial prediction based noise reduction technique (MSP-BF), and discussion about the relationship between MSP-BF and MWF. In MSP-BF, noise sources can be reduced effectively at the first stage by multichannel spatial filtering. The desired sources are distorted after the first stage. The distortion of the desired sources are restored at the second stage by multichannel distortion-restoration filtering. The advantageous point of MSP-BF is that this method can reduce noise sources without the steering vectors of the desired sources. However, there is not a sufficient theoretical analysis in the previous paper. The second contribution is that an alternative noise reduction technique based on the subspace filtering at the first stage, NSR-BF, is proposed. The experimental result under a reverberant environment is shown.