The ESCA Workshop on Speech Synthesis September 25-28, 1990 Autrans, France

Automatic Linguistic Processing in a German Text-To-Speech Synthesis System

Betina Schnabel (1,2), Harald Roth (1)

(1) Centre National d'Etudes des Telecommunications Lannion, France (2) Institut de la Communication Parlee Grenoble, France

The linguistic processing of the CNET's Text-to-Speech (TTS) synthesis for German has recently been automatized and extended by a series of new algorithms. The modifications concern the following three stages: first, a preprocessing stage was added to convert numerical expressions, abbreviations and diacritics into sequences of graphemes. Then, the morphological analysis has been extended to compound words - one of the major problems of German morphology - and to the grammatical labelling of word classes. Finally, an automatic parser has been developed for the insertion of syntactic-prosodic markers. It uses a set of hierarchized parsing rules which are applied to the extracted sequences of grammatical categories resulting from the upper analysis.

Full Paper

Bibliographic reference.  Schnabel, Betina / Roth, Harald (1990): "Automatic linguistic processing in a German text-to-speech synthesis system", In SSW1-1990, 121-124.