An improved spectral subtraction method based on modified MCRA noise estimate and magnitude compensation

Conventional spectral subtraction algorithm is limited by its stationary noise assumption. However, most of the nonstationary noises are encountered in real world. In this paper, a novel refined spectral subtraction algorithm for speech enhancement is proposed. The algorithm based on both MCRA (Minima Controlled Recursive Averaging) noise estimate algorithm and magnitude compensation algorithm, which could keep residual noise and speech distortion under a lower level.