Fourth ISCA ITRW on Speech Synthesis
August 29 - September 1, 2001
FIPSVox is a text-to-speech system for French developed at LATL. It is based on FIPS, a large-scale, multi-purpose, GB-based syntactic parser which produces detailed analyses and the MBROLA diphones-concatenation synthesizer. The syntactic information provided by the parser is directly exploited by the grapheme-to-phoneme module to handle heterophone homographs as well as French elision, denasalisation and liaison phenomena. The prosody generation module also uses this information to determine the dependency between phrases, the accentuation of syllables, and to identify particular syntactic structures such as extraposed constructions (cleft, heavy-NP shift, left- dislocation structures, etc.), and parentheticals to derive of appropriate prosodic patterns.
Bibliographic reference. Goldman, Jean-Philippe / Gaudinat, Arnaud / Nerima, Luka / Wehrli, Eric (2001): "FipsVox: A French TTS based on a syntactic parser", In SSW4-2001, paper 205.