Performance evaluation of nonlinear speech enhancement based on virtual increase of channels in reverberant environments

In this paper, we evaluate the performance of a maximum signal-to-noise ratio beamformer based on a virtual increase of channels. We previously proposed a new microphone array signal processing technique, which virtually increases the number of microphones by generating extra signal channels from two real microphone signals. This technique generates a virtual observation on the assumption that the sources are W-disjoint orthogonal, which means that only one source is dominant in one time-frequency bin. However, mixed signals with a long reverberation tend to dissatisfy this assumption. In this study, we conducted experiments in a variety of reverberant environments, as well as computer simulation using image method. As a result, we confirmed that our technique contributes improving the performance in reverberant environments. We also confirmed that the longer the reverberation time, the smaller the increase in the improvement using our technique. Moreover, we present directivity patterns to confirm the behavior of a virtual increase of channels.