An effective model of stressing in an Arabic Text To Speech System

The improvement of the naturalness a TTS synthesis system implies the development of a prosodic model which is able to reproduce, from a text, the acoustic phenomena implied in the production of the natural word. This requires formalism which interpret the prosody of a written text and a phonetic model which allows the quantification of prosody so that it is exploitable by the synthesis system. The object of this article consists in treating the intonation under the aspect of the lexical stressing with the level of the words, the simple sentences and the texts. A model is proposed and a first evaluation in the ARAVOICE system is presented.