This paper gives a first evaluation of the rule-based Text-to-Speech system for French which has been developed within the ESPRIT Polyglot project 1024 "A Multilingual Speech-to-Text and Text-to-Speech system". The three modules which process the phoneme-to-parameter conversion are reminded with an emphasis on the coarticulation rules. As the evaluation concerns the segmental level, three identification tests are carried out and proposed to ten listeners. They consist of isolated vowels, VCV and VCCV logatoms. The results show that the weakest points are the nasal vowel / /, the voiced fricatives with /v/, in particular, and the liquid /l/. Further improvement is also needed to model the coarticulation in the V-/u/ and /u/-V contexts.

Keywords: Text-to-Speech system, rule synthesis, identification tests.