Delayless speech enhancement with a virtual zero-phase response using a prediction of periodic signal components

In this paper, a delayless speech enhancement scheme with zero phase distortion is proposed. It is based on a cascade of adaptive filters that predicts periodic components with a significant auto-correlation for lags larger than a value D. The adaptive filter is positioned at the output of a speech enhancement algorithm, to adjust the phase of the periodic components to the noisy signal, and to remove stochastic signal components with a significant auto-correlation only for lags smaller than D. The stochastic components are enhanced in a separate channel and mixed back together with the periodic components to give an output with no delay or phase distortion compared to the input signal. Such a scheme is useful for low-delay processing where the phase of the signal must be preserved, for instance as a front-end for spatial filtering or when the output is mixed with another source, such as the direct transmitted sound through the vent in an open hearing aid fitting.