Natural Language Generation for Acoustic Output

This project laid the foundations for a system for the concept-to-speech system VieCtoS that combines natural language generation with speech synthesis. The project focussed on architectural issues, especially on laying the foundations for a representation that combines syntactic and semantic structures allowing for the derivation of prosodic features relevant for better intonation.

On the speech synthesis side the prototype of a demi-syllable based synthesizer originally developed at the Institut für Nachrichtentechnik und Hochfrequenztechnik was extended and adapted. This included extending and improving the existing inventory and developing a suitable interface to the (phoneme-based) generator.

On the natural language generation side an existing generator (developed at OFAI on the basis of FUF) was coupled with a phoneme-based lexicon and a grammar suitable for the generation of spoken utterances.

Research staff

• Ernst Buchberger
• Johannes Matiasek
• Hannes Pirker
• Harald Trost

Partners

• Institut für Nachrichtentechnik und Hochfrequenztechnik (TU Wien)

Sponsor