EUROSPEECH 2003 - INTERSPEECH 2003
In this paper, we present a multi-domain spoken dialogue system equipped with the capability of parallel computation of speech-recognition engines that are assigned to each domain. The experimental system is set up to handle three different domains (restaurant information, weather report, and news query) in an in-car usage. All of these tasks are of information retrieval nature. The domain of a particular utterance is determined based on the likelihood of each speech recognizer. In addition to the human-machine interaction, synthesized voice of the route sub-system interrupts the dialogue frequently. Experimental evaluation has yielded 95 percent recognition accuracy in selecting the task domain based on a specially designed scoring method.
Bibliographic reference. Isobe, T. / Hayakawa, S. / Murao, H. / Mizutani, T. / Takeda, Kazuya / Itakura, Fumitada (2003): "A study on domain recognition of spoken dialogue systems", In EUROSPEECH-2003, 1889-1892.