Tapping-noise suppression with magnitude-weighted phase-based detection

This paper proposes tapping noise suppression with a new phase-based detection. Phase slope of the input noisy signal is compared with an ideal phase slope obtained from an average of intra-frame slopes along the frequency axis. In order to cope with heavily low-pass characteristics of tapping noise spectrum, phase values are weighted with the magnitude at each frequency point. Phase unwrapping problem is alleviated by use of a rotation vector of frequency domain components. Comparison of enhanced signal spectrogram with that of clean speech demonstrates superior enhanced signal quality.