A multi-channel postfilter based on the diffuse noise sound field

In this paper, we present a multi-channel Directional-to-Diffuse Postfilter (DD-PF), relying on the assumption of a directional speech signal embedded in diffuse noise. Our postfilter uses the output of a superdirective beamformer like the Generalized Sidelobe Canceller (GSC), which is projected back to the microphone inputs to separate the sound field into its directional and diffuse components. From these components the SNR at the output of the beamformer can be derived without needing a Voice Activity Detector (VAD). The SNR is used to construct a noise cancelling Wiener filter. In our experiments, the developed algorithm outperforms two recent postfilters based on the Transient Beam to Reference Ratio (TBRR) and the Multi-Channel Speech Presence Probability (MCSSP).