A spoken language interface to interactive multimedia services

In the current generation of interactive multimedia services (IMS), the primary means of user input is by pushing buttons on a remote control handset, as in BT´s Interactive TV trial in the Ipswich and Colchester areas, or by touching icons on a screen, as in BT´s current Touchpoint kiosk trial. If the service has a fairly small amount of information, organised in manageable hierarchies of menus, this may be a satisfactory interface, but as the service grows in size and complexity, the user is likely to have to make a long, complex or tedious series of button pushes to navigate to the information he desires. If spoken language is added as an input modality to the interface, it becomes much more powerful, enabling the user to go directly to the information required. This paper describes a demonstrator system called Pik-a-Pic!, which provides a spoken language interface to a small subset of the Internet Movie Database. A user can query the system, using continuous speech, for films according to title, actors or director, separately, or by a combination of actor and director. The system will create and display a WWW page with the information requested. This page also contains links to the relevant pages of the Internet Movie Database itself. The paper investigates the combination of language models within the recogniser for robust performance, along with the benefit of using caseframe parsing as a means of robust natural language analysis