Managin speaker identity and user profiles in a spoken dialogue system

In this paper we present an approach to capture user preferences dur- ing their interactions with a spoken dialogue system (SDS) and to exploit these preferences into the dialogue manager module. We use the output of an unsuper- vised speaker identification module and a natural language understanding module to adapt the behaviour of the SDS to each particular user by developing a statistic analysis of the preferred goals of each user. This way we can adapt the behaviour of the system to each user's preference, anticipating the goals that a user wants to fulfill. We have evaluated the speaker identification system with two different databases, and we have obtained identification errors below 4%. The initial evalu- ation of the profile manager shows an improvement on subjective metrics, such as users' perception of efficiency or naturalness of the system responses.