Romanian Prosody Prediction Module Based on a Functional Intonational Model

This paper presents a prosodic prediction module used by the Romanian Text-to-Speech (TtS) system in intonation synthesis. The prosody prediction refers to the surface generation of the F0 contour, based on the F0 patterns assigned to the functional categories of the prosodic units. Prior to the prediction module presentation, the paper includes a summary of these functional categories and the partial melodic contour descriptions based on functional labels. The block diagram of the prediction module outlines two main processing steps: the phrasing prediction for building the utterance tree and the selection of the melodic contours of its groups. Both processing steps are exemplified within a case study of Romanian text speech synthesis. The prosody prediction results are discussed and compared with natural F0 contours of different speakers.