Spoken Language Processing Group (TLP)

Groupe TLP - RAILTEL Project

The spoken language system runs a Unix workstation with a high quality telephone interface which can support up to 4 telephone lines. The spoken query is decoded by a speaker independent, continuous speech recognizer. The speech recognizer has been ported to deal with telephone quality speech. The recognizer output is passed to the natural language understanding component. In our current implementation the output of the speech recognizer is the best word sequence, however, the recognizer is also able to provide a word lattice. The semantic analyzer carries out a caseframe analysis and generates a semantic frame representation.

The dialog history and default values generated from task knowledge are used to complete missing information in the semantic frame. The dialog manager prompts the user to fill in any additional missing information necessary for database access and then generates an SQL-like query to the information database. The system accesses a copy of the static train information (database RIHO) via a network connection. The returned information is then converted to a natural language response, which is synthesized by speech concatenation of units in a pre-stored dictionary and played to the user.

The spoken language system allows a mixed-initiative dialog where the user can provide any information at any point in time. Experienced users are thus able to provide all the information needed for database access in a single sentence, whereas less experienced users tend to provide shorter responses, allowing the system to guide them.