Project description

A spoken dialog interface will need to contain a robust speech recognizer for transforming the (noise corrupted) user speech to text; a language understanding component which extracts the task relevant information from the recognized word sequence (which may contain erroneously recognized words); a dialog management system generating textual prompts for missing information information verification, greetings and answers to queries; and a speech output system that converts the textual prompts to spoken messages.

The project will develop the tools and competence that constitute the fundament for the development of future spoken dialog systems. In addition, the efforts within the project sub-tasks will enhance the present state-of-the-art within those areas. The collaboration between the three university departments shall result in a coordination of existing and the definition of new courses, particularly at a post-graduate level, yielding a comprehensive educational package in the area of spoken language engineering.

The general goals of the project shall be obtained through the development of a prototype system for public transport information. The system will be specified in cooperation with Trafikanten MidtNorge and will be targeted as a telephony based automated public transport information service aimed at the general public.

The project is organized as a collaborative effort between NTNU and SINTEF. A significant amount of the research will be performed in the form of 3 doctoral fellowships in Part I and an additional 2 in Part II, and through student theses within the framework of the project.

The project will have strong collaborative ties with Telenor R\&D, both within speech input/output, language resources and natural language processing. Telenor will provide access to their existing technology for research purposes in exchange for access to project results. In addition to being on the project steering committee, Telenor R\&D will also participate in ordinary project meetings on an ad-hoc basis.


For further information contact Professor Torbjørn Svendsen
Last modified: Tue Oct 21 15:41:52 MET DST 1997