Computers that look, sound and respond like people have been the stuff of science fiction. Now Ian Grayson sees them coming to life at BT Labs
In the cult television series Red Dwarf, the crew members are assisted in their deep space adventures by a computer called Holly. Holly appears as a talking head on a screen, communicating with the crew through speech and facial expressions. If you didn't know better, you'd think Holly was a real person. For those battling daily with a keyboard and mouse, the idea of communicating with a computer that looks, sounds and responds like a human being seems a long way off - but it may not be as far into the future as you think.

Staff at the BT's research and development labs in Suffolk have been working with talking heads for almost 10 years. During that time, their creations have progressed from sluggish, unconvincing images needing vast computational power, to unnervingly realistic depictions of real people that will run on a desktop PC.

BT's latest three-dimensional talking heads can adopt the facial features and voice of any person. They can, smile, frown, nod, turn, and even show anger.

For Andy Breen, a BT researcher, the idea of a computer that communicates through a human-like interface is an enticing prospect. "What we really want is to present a personality with a voice and a face that give the user the perception of having someone talking to them," he says. "What we're trying to generate is a synthetic persona."

BT's interest in synthetic personas began in the early Eighties. At that time, the challenge was to find a way of sending good-quality video images down conventional telephone lines. To reduce the amount of information needing to be transmitted, researchers came up with the idea of sending just one image, followed by controls that modified elements such as lip movements and facial expressions. Though a partial success, the resulting face was somewhat jerky and unrealistic.

The current generation of talking heads has come a long way. Each is based on a three-dimensional wire frame that is changed to match the facial features of a particular person. The frame is overlaid with a digital image of the person's face, giving the illusion of a three-dimensional head.

Researchers have found that there are 13 key muscles in the human face that combine to produce most everyday expressions. The wire frame models have been designed to alter the digital facial image in the same way, to produce different expressions.

Users enter text via a keyboard that is converted into spoken words by BT's Laureate software, a sophisticated conversion program that converts text to realistic speech in a variety of accents.

The head uses 19 different mouth models to simulate the shapes produced during normal speech. Each model incorporates lip, teeth, jaw and tongue positions. The models are blended together using morphing techniques to make the resulting moving image look as natural as possible. This is achieved by selecting a slightly different wire frame model for each sound.

The stages between each are then generated and displayed in turn, giving the impression of fluid movement. The end result is similar to a film being projected at around 10 frames per second.

"We're not just showing a cartoon," says Breen. "It's not pre-done and then displayed - it's being generated in real time by the computer."

Where once this type of application would have taken a high-powered work station, the talking head can now be run on a standard desktop PC. But as processing power has increased, the goals of developers have become more ambitious. They are now looking at ways of creating entire virtual worlds in which people can meet and interact.

"A person's image is captured by a camera, translated into computer information and then regenerated inside a virtual world," explains Breen. "The application here is virtual conferencing, where you don a headset and have a meeting with people around a table even though they are not actually there."

Early examples of such technology can already be seen on the World Wide Web. Sites have been established where users enter an online "world", adopt a character and move around interacting with other users. BT's talking- head technology will take this concept even farther, perhaps allowing people to use their own faces.

But for Breen and his team of researchers, the most challenging aspect of the technology is the development of an artificial persona - a "Holly" - that improves the way people interact with computers.

"I'm concentrating on giving a computer a character, its own emotion," Breen says. "When I have a conversation with a computer I want to be able to get the perception of talking to a real person."

Breen believes the first 3D talking heads will begin to appear soon, on Web sites or as characters in computer games. Widespread use should occur within two years.

Future applications of the technology could include electronic personal secretaries that read e-mail and respond to user requests, and virtual guides that trawl for information and then present it to the user in spoken form.

Holly may still be fictional at present, but if BT's researchers have their way, she may be coming to a PC near youn

BT Labs