Wiener filtering with spectrum estimation by wavelet transformation

Current methods used to improve the signal-to-noise ratio mainly employ Wiener filtering, or methods derived from it, such as spectral subtraction. All these methods assume that it is possible to determine or at least to estimate noise spectral characteristics. As can be derived, the estimation by the periodogram is not exact, but it contains a disturbance. Averaging the power spectra offers better results but the properties are simultaneously downgraded by a non-stationary disturbance. That is why the paper deals with the improvement of the power spectral density (PSD) estimation. To achieve this we use the method of thresholding wavelet-transform coefficients, which we apply to each periodogram separately. Then the resultant estimation is used for the Wiener filter. The differences between the estimations obtained by the periodogram, by averaging and by this new method are shown. When we use the wavelet transformation, a marked improvement in suppressing the disturbing signal appears. As a result, we obtain a good noise reduction while preserving a good response to changes in the nature of the disturbing signal.