We introduce a method to automatically evaluate the weights of the digital filter used for high-resolution spectroscopy in a mixed analog-digital setup, which operates with minimum human attendance. The optimum filter weighting function WF is obtained from the noise autocorrelation; an ultra-accurate estimate of the singularities of the antialiasing filter is derived from its experimental pulse response. From these data the procedure automatically computes the optimum WF and the digital filter weights. We show that the method provides a much better flatness of the flat top (to within 0.1% of the peak value); a more precise elimination of tails in the WF (to better than 0.1% of the peak value) and a much lower quantization noise (more than a factor 10) at the filter output than other possible methods (DFT or Wiener algorithms). It was successfully tested in the generation of trapezoidal and optimum cusp-like WFs even in presence of non-negligible noise. This calibration procedure runs in a few seconds with no additional hardware.