Inversion of F0 Model for Natural-Sounding Speech Synthesis