The advances of speech processing technologies make it possible to build spoken dialogue systems (SDS) which may provide people with useful information. However, most current SDS can only deal with one speaker a time, improvements should be made before the SDS can be applied to domains where multiple speakers interact with the system. This paper discusses research issues for developing a multi-speaker dialogue system (MSDS) which is able to retrieve various mobile information in the car environment.

The differences between traditional (single speaker) and multi-speaker dialogue system are first addressed. Then, two research topics are studied. 1) Speech source identification, which determines the active speaker. 2) Multi-speaker dialogue management, which interpreter the intention and maintain the dialogue history of the speakers to keep the interaction smooth.

Many testers in a car environment attended the experiment for active speaker detection and multi-speaker dialogue system. The experiments showed an encouraging result that the proposed approach of MSDS did work properly.