The method of fractal interpolation of functions is used, that is, a digitized speech signal waveform is approximated with a piecewise self-affine fractal interpolation function. In this method, a speech frame is divided into larger pieces called domains, and smaller ones called regions, and an optimal set of affine transformations of domains into regions is calculated. The greater are the regions and domains, the better is the compression ratio, though the quality of the restored signal may be worse.