Multimodal systems for children: building a prototype

In this paper, we describe our e orts in designing and building a prototype multimodal system for children users. Data collection e orts and user experience results from a WoZ study using a popular computer game are reviewed rst. Automatic speech recognition and spoken language understanding technology for children speakers are discussed next. A multimodal prototype is designed for a personal agent and a gaming application. Emphasis is placed on a modular architecture, handling of multimodal input and multimedia output, and providing an engaging user interface. Informal evaluation by children users was very positive especially for the animated agent and the speech interface.