Prosody control in fluent Dutch text-to-speech

Fluent Dutch Text-To-Speech is a new, high-quality TTS system for Dutch, built on top of the MBROLA diphone synthesizer (Dutoit 1997), and commercially available for Windows 95/NT. This paper discusses two applications which use the TTS system: a talking dictionary, and a communication aid for the vocally handicapped. In the talking dictionary application, parameterized prosody rules enable the generation of various pronunciations of a word, from formal to informal. For the vocally handicapped, prosody controls are used to create different voices and moods. 1. TALKING DICTIONARIES In collaboration with Van Dale Lexicografie bv, a Dutch dictionary publisher, we created a version of our system for use in cdrom-based dictionaries. In this application, isolated words are synthesized from phonemic transcriptions which are part of the dictionary. In contrast with full text-to-speech conversion, as the pronunciation of a word is already known, the text analysis front end of the TTS system is not used. The transcription system developed by Van Dale encodes phonemic information, as well as syllabification and stress. In addition, information is included about: • optional sounds, deletions and insertions