Single Channel Speech Enhancement Based on Perceptual Frequency-Weighting

Speech Enhancement based on Sequential Noise Estimation with a Masking Property

“…The well-known basic method of speech enhancement is based on spectral subtraction [1,2], which tries to calculate the spectral speech magnitude by subtracting a noise component estimated from the noisy speech and then combines this calculated spectral speech magnitude with the phase of the noisy speech. In previous papers, we proposed an approach to speech enhancement that was based on the introduction of a perceptual criteria using a frequencyweighting filter in a subtractive-type enhancement process [6,7,9]. Our previous presented paper suggested that "the noise elements are considered to be masked by the speech power in the formant regions, and, conversely, unmasked in the valleys between the formants.…”

“…Our previous presented paper suggested that "the noise elements are considered to be masked by the speech power in the formant regions, and, conversely, unmasked in the valleys between the formants. Therefore the gain factor, which decides the amount of estimated noise subtracted from the noisy speech, is controlled to be lower in the formants and higher in the valleys" in 2007 [7], referred by Chowdhury, F.A. et al in their paper [8].…”