Faculty
Takeshi Kawabata
Major research activity
Speech communication is a promising medium for constructing an effortless human-machine information interchange system. However, up-to-date speech recognition technologies could not achieve the practical recognition accuracy for this purpose yet. On the contrary, nearly 100% task completion is accomplished in human-to-human communication by applying dialog coordination and confirmation behaviors. Therefore the "Spoken dialog system" is one of the most human-friendly computer interfaces.
Traditional dialog systems cost a lotto prepare sophisticated dialog scenarios for each task. A difficulty is that the scenarios have to include very complex dialog flows to cope with the various dialog behaviors of human. In our architecture, a whole dialog scenario is written as a tree of micro-dialog-scenarios (MDS). If an incoming speech token cannot be handled by a current MDS, the token is passed to the parent node and retreated by the MDS of the parent node.
A real (or virtual) humanoid robot is an application of the spoken dialog system. The robot has to recognize not only the speech content, but also its incoming direction. The signal processing technologies for detecting speech direction under the severe mechanical noise conditions are researched. We also investigate the feasibility of the complex system theory to the signal processing fields.
Major relevant publications
- Kinoshita, Y. and Kawabata, T.: "Timeout Controls of the Voice Command Context for Telematics Systems," IPSJ SIG Tech. Rep., Vol.2014-SLP-104, No.18, pp.1-6 (Dec. 2014)
- Ohshima, H. and Kawabata, T.: "Consideration about the Sparseness of Parameter Initialization of PLSA Language Models," IPSJ SIG Tech. Rep., Vol.2013-SLP-97, No.6, pp.1-5 (July 2013)
- Higashinakagawa, R. and Kawabata, T.: "Dependency between the Survival Density and Initial Configuration of the Stochastic Game of Life," Tech. Rep. IEICE, NLP2009-7-36, pp.1-6 (July 2009)
- Kawabata, T., Fujiwara, M. and Shibutani, T.: "Detection of Speaker Direction based on the On-and-Off Microphone Combination for Entertainment Robots," Entertainment Computing - ICEC2005, Lecture Notes in Computer Science, Springer, LNCS-3711, pp.248-255 (Sep. 2005)
- Kawabata, T.: "Object-Oriented Dialog Management Architecture for Designing Multi-Purpose Spoken Dialog Systems," The 18th International Congress on Acoustics (ICA-2004) (Apr. 2004)