Listen first: Output-based multi-microphone speech enhancement

Panos Apostolidis, Svend Feldt, Zheng-Hua Tan +2

Traditionally, hearing-aid speech enhancement (SE) algorithms rely on input-based feature estimation, often derived by a voice activity detector (VAD), to configure beamformers. Ye…