DMANET: DEEP LEARNING-BASED DIFFERENTIAL MICROPHONE ARRAYS FOR MULTI-CHANNEL SPEECH SEPARATION

In this paper, we develop a novel differential microphone ar-rays network (DMANet) for solving the multi-channel speechseparation problem. In DMANet we explore a neural networkcombined to differential microphone arrays (DMAs) beam-forming technique. Specifically, a sequence of differentialoperation is introduced alternately into network. Based onthe filter-and-sum network (FaSNet), we show how DMANetsignificantly improves the separation performance. Numeri-cal experiments demonstrate that the proposed network has aclearly advantageous improvement on SI-SNR with a smaller model.