I. Introduction and Relation to Prior Art
Acoustic noise reduction is vital to hands-free and hand-held speech communication [1], [2], hearing aids [3] or true-wireless stereo headphones. Typical single-channel algorithms include the spectral-subtraction [4] and numerous minimum-mean-square error approaches mostly operated with decision-directed a-priori SNR [5]–[8]. From the range of multiple-input/single-output algorithms, the minimum-variance distortionless response beamformer and the multichannel Wiener filter are frequently used for speech enhancement [9]. The subclass of binaural algorithms for hearing -aids derives with a further constraint for spatial-cue preservation [10]–[14].