Prosody modification and Fujisaki's model: Preserving natural soundness

Control of prosodic characteristics is one of the most important problems in the area of speech synthesis. Fujisaki's model is probably the best model for pitch variations and its inversion is suitable for being integrated within speech synthesizres. This paper proposes a speech synthesis method based on Fujisaki's model (combined direct and inverse modeling) in order to preserve natural soundness of synthesized speech. The idea is to modify a pitch contour on the basis of Fujisaki's features and a reference contour. Experimental results have shown that using constraints related to Fujisaki's model guarantees good natural-sounding speech synthesis.