Talking heads exercise

Avatars, help agents, and other graphical talking heads are of increasing interest in commercial products and educational applications. Many non-technical people cannot differentiate between a rendered, prestored video with speech, and a live conversational agent that synthesizes the output in real time based on user input. The purpose of this exercise is to show you some tools for building talking avatars, and for you to judge which ones might be usable for conversational agents, and what domains they might be appropriate for.

For starters, check out the following two dialog agent demos:

The Medadvisor demo movie from UofR to see the interaction behaviors implemented in the talking pill.
The Intelligent Maptask agent on your flash drive.

The tools to be evaluated are:

Veepers: Play with them in the American Greetings online demos, or play with a similar technology on seestorm.com
Sitepal (click on Demo it and use the 'add audio' feature). A note about the interface for Sitepal. Even when you purchase the tool, the tool for creating characters and audio files is embedded in a web interface.
Facegen: see the demo video on your flash drive
MikeTalk: you can't make your own videos, but you can play some previously rendered videos at the web site. Also check out Mike's cousin Mary 101

These tools differ in the amount of photo-realism of the avatar, the quality of synchronization between the lip movements and audio output, the ability to plug the avatar into a running application, the ability to create non-human characters, and the ability to create gestures to accompany the dialog, among other things.

In your group, imagine that you are a software consultant evaluating some avatar software that might potentially be included in a variety of dialog systems:

A bus schedule information kiosk at a city bus terminal
A pediatric medical information database to be installed on home computers
A fun pop-up help agent that pops up in a separate pane (like the Microsoft paperclip) that is part of a children's drawing tool

For each application, what characteristics do you think a talking character should possess? Do you need good pausing/parking behaviors, good turn-taking behavior to distinguish listening posture from speaking postures, variable eye-gaze, human vs. non-human characters, etc? In your group, come up with three or four desiderada you would use to choose an avatar for a particular application.

donna byron

Last modified: Wed May 30 01:51:57 EDT 2007