5th International Conference on Spoken Language Processing
In this paper we present a speech understanding system that accepts continuous speech sentences as input to command a HIFI set. The string of words obtained from the recogniser is sent to the understanding system that tries to fill in a set of frames specifying the triplet (SUBSYSTEM, PARAMETER, VALUE). All circumstances (understanding incompleteness, HIFI set status, result of the command execution) are confirmed back to the user via a text to speech system with substitutable-concept pattern-based generated messages. The understanding engine is based on semantic-like tagging, including "garbage" tag, and context-dependent rules for meaning extraction. The system allows the application developer to follow the reasoning process (as every understanding rule has an associated concept pattern), spoken by the speech generation module. The concepts for speech generation are randomly substituted with alternative expressions having the same meaning to achieve a certain degree of naturalness in the response speech.
Bibliographic reference. Ferreiros, Javier / Colas, Josť / Macias-Guarasa, Javier / Ruiz, Alejandro / Pardo, Josť Manuel (1998): "Controlling a HIFI with a continuous speech understanding system", In ICSLP-1998, paper 0988.