Unit selection and its relation to symbolic prosody: a new approach

Abstract

Aiming at the improvement of the quality of synthetic speech generated by our TTS ARTIC, we adopted the unit selection method. Our unit selection module is driven by prosody described solely by high-level symbolic features which are linked to the prosody of synthesized phrases through the phenomena of prosodic synonymy and homonymy. It was confirmed that such an approach generates speech with high naturalness and keeps the richness of prosody. The quality of the output speech generated by our approach was assessed as very close to natural. The concept of prosodic synonymy and homonymy is, therefore, further extended and formally described in this paper, and its importance to the unit selection treatment is demonstrated. The difference of this concept from the concepts most frequently used is also shown. Moreover, the first experiment following the formal definition of the problem presented in this paper has been carried out.