Multi-Speaker Dialogue for Mobile Information Retrieval

Currently, most Spoken Dialogue Systems (SDS) only deal with the interaction between the system and one speaker. In some situations, interaction may occur between several speakers and the system. This paper proposes methods for the Multi-Speaker Dialogue System (MSDS) which allows user to retrieve useful information such as navigation guide, weather forecast, etc., in the car environment. The differences between traditional SDS and MSDS are addressed first. The interaction between speakers and the MSDS are classified into three types i.e., independent, cooperative, and conflict. Then, we focus on two major research topics of the MSDS, i.e., speaker source identification which determines the active speaker and multi-speaker dialogue management which interpreters the speaker intention and maintains the dialogue histories to keep the interaction goes smoothly. . Twenty-four testers attended the experiment for active speaker detection and multi-speaker dialogue system. The experiments showed an encouraged result that the proposed approach works properly, and it provides user-friendlier interface for multi-speaker interaction in the car environment.