Automatic User Classification for Speech Dialog Systems

Abstract

The usability of automatic speech dialog systems depends – beside a robust speech recognition – on a target user adequate dialog design. One way to enhance the usability of such human-machine interaction is to design adaptive systems. Dealing with a high number of heterogeneous user types, a speech dialog system adapting to the respective user type is believed to receive high acceptance rates. But since many speech dialog systems are used anonymously, there is only little previous knowledge about the type of user calling. Therefore, information has to be gathered automatically by the speech dialog system itself. In our study we focused on log files as a source of information. Created during running dialogs, log files contain detailed entries about events of the dialog with precise timestamps. We illustrate the chances of automatically gaining information about the user as a contribution to a more user centered dialog.