Genetic optimization of piecewise self-affine fractal interpolation with application to speech modeling