XPSpeech ("Verbesserung der Benutzerschnittstelle von Expertensystemen durch den Einsatz gesprochener Sprache") is a Speech Interface to Expert Systems that has been developed together with the Institut für Nachrichtentechnik und Hochfrequenztechnik at the Vienna University of Technology. In this project, the application of natural language processing and speech techniques to improve the user interface of expert systems was investigated. Possible application areas for such a system are domains where the user's vision is occupied by some task and where the dialogue is mostly system-driven. The system was implemented in the domain of electronic circuit repair that fulfils the necessary preconditions in an ideal way.

The expert system supported the user in diagnosing and repairing the error by producing a sequence of actions (checking and exchanging parts) to be performed by the user. A user model translated these actions into more or less explicit chain of requests depending on the assumed skills of the user.

These were translated into natural language utterances by a generator. Special emphasis was put on an adequate description of components (e.g, anaphoric references) and on providing focus information. In a last step these utterances were acoustically realized by a formant synthesizer. Feed-back to the system by the user was restricted to a simple set of notifications like "yes", "no", "done", "explain".

Duration: 1987 - 1989
Sponsor: Jubiläumsfonds der Österreichischen Nationalbank
Researchers: Ernst Buchberger, Georg Dorfner, Johannes Matiasek, Harald Trost
Partners: Institut für Nachrichtentechnik und Hochfrequenztechnik (TU Wien)