Noise power spectrum estimation based on weak speech protection for speech enhancement

Noise power spectrum estimation is a fundamental component of speech enhancement. In the paper, an improved noise power spectrum estimation algorithm and weak speech protection smoothing are proposed for non-stationary noise environments. The estimated noise is obtained as an average of the weighted noisy speech using a constant smoothing parameter, the weighting factor is adjusted by an estimated signal-to-noise ratio (SNR), and the SNR is controlled by the weak speech protection smoothing scheme with a time-frequency dependent protection smoothing factor. The new proposed weak speech protection smoothing scheme can avoids noise overestimates in the weak speech segments following high SNR speech, and makes more accurate SNR estimate. Objective experimental results and a subjective comparison show that the improved noise estimation algorithm when integrated in speech enhancement is preferred over the competitive noise estimation algorithms.