SD-TEAM: Interactive Learning, Self-Evaluation and Multimodal Technologies for Multidomain Spoken Dialog Systems

Abstract

Speech technology currently supports the development of dialogue systems that function in limited domains for which they were trained and in conditions for which they were designed, that is, specific acoustic conditions, speakers etc. The international scientific community has made significant efforts in exploring methods for adaptation to different acoustic contexts, tasks and types of user. However, further work is needed to produce multimodal spoken dialogue systems capable of exploiting interactivity to learn online in order to improve their performance. The goal is to produce flexible and dynamic multimodal, interactive systems based on spoken communication, capable of detecting automatically their operating conditions and especially of learning from user interactions and experience through evaluating their own performance. Such ?living? systems will evolve continuously and without supervision until user satisfaction is achieved. Special attention will be paid to those groups of users for which adaptation and personalisation is essential: amongst others, people with disabilities which lead to communication difficulties (hearing loss, dysfluent speech, ...), mobility problems and non-native users. In this context, the SD-TEAM Project aims to advance the development of technologies for interactive learning and evaluation. In addition, it will develop flexible distributed architectures that allow synergistic interaction between processing modules from a variety of dialogue systems designed for distinct tasks, user groups, acoustic conditions, etc. These technologies will be demonstrated via multimodal dialogue systems to access to services from home and to access to unstructured information, based on the multi-domain systems developed in the previous project TIN2005-08660-C04.