New IEEE/ACM Journal Article on Nonlinear Beamforming
18 April 2021, by David Mosteller
Our paper "Nonlinear Spatial Filtering in Multichannel Speech Enhancement" by Kristina Tesch and Timo Gerkmann has been accepted for publication in the IEEE/ACM Transactions on Audio, Speech, and Language Processing. The preprint is available here.
The paper shows via theoretical and experimental analysis that in non-Gaussian noise, the state-of-the-art concatenation of a linear beamformer and a postfilter is not optimal. The paper thus suggests to do further research into nonlinear spatial processing by deep neural networks.
Impressive audio examples are given here.