7th International Conference on Spoken Language Processing
September 16-20, 2002
A speech denoising technique based on subband noise estimation and a perceptual modification of Wiener filtering is proposed. The noisy speech is first decomposed into critical band signals by an auditory filterbank and the denoising is carried out on the subband signals. The time varying subband noise variance required for denoising is estimated by tracking the minimum variance of the subband noisy signal. Auditory masking is applied to calculate the optimum denoising gain using a Wiener filtering approach to minimize the perceived speech distortion. This denoising technique is suitable for coloured and non-stationary noise environments. It gives natural sounding speech with a very low level of musical noise.
Bibliographic reference. Lin, L. / Holmes, W. H. / Ambikairajah, E. (2002): "Speech enhancement based on a perceptual modification of wiener filtering", In ICSLP-2002, 781-784.