Noise estimation by minima controlled recursive averaging forrobust speech enhancement

摘要

In this letter, we introduce a minima controlled recursivenaveraging (MCRA) approach for noise estimation. The noise estimate isngiven by averaging past spectral power values and using a smoothingnparameter that is adjusted by the signal presence probability innsubbands. The presence of speech in subbands is determined by the rationbetween the local energy of the noisy speech and its minimum within anspecified time window. The noise estimate is computationally efficient,nrobust with respect to the input signal-to-noise ratio (SNR) and type ofnunderlying additive noise, and characterized by the ability to quicklynfollow abrupt changes in the noise spectrum