Wiener Filters for the Reconditioning of Old Speech Recordings Under Uncertainty

The article revisits the design of optimal Wiener filters under uncertainty, specifically under limited information and imprecision of the available data on the noise and signal spectra, as in case of old speech recordings. The derivation of the filters is based on a few reasonable assumptions regarding the noise measurement and on the variability of the noise. The filters use several properties of the speech signal in the choice of the filter structure and optimization procedure.