Single-channel speech enhancement method based on masking properties and minimum statistics

A single-channel speech enhancement method of noisy speech signals at very low signal-to-noise ratios is presented, which is based on masking properties of the human auditory system and power spectral density estimation of nonstationary noise. It allows for an automatic adaptation in time and frequency of the parametric enhancement system, and finds the best tradeoff among the amount of noise reduction, the speech distortion, and the level of musical residual noise based on a criterion correlated with perception and SNR. This leads to a significant reduction of the unnatural structure of the residual noise. The results with several noise types show that the enhanced speech is more pleasant to a human listener.