Fractal modeling of fluctuations in the steady part of sustained vowels for high quality speech synthesis