Virtual humans edge closer - Tilted Forum Project Discussion Community

Tandem · 07-27-2003, 07:13 PM

By Spencer Kelly
BBC

For years, one of the main goals in computer graphics has been to recreate a totally convincing human being on screen, something that looks and acts so life-like that it is indistinguishable from a real person.

But human beings are hard to recreate, because along with all those tiny details in the way we look, there is the way we move, our expressions.

Now there is a new type of computer graphics card which is bringing virtual humans a lot closer.

Dawn Fairy is a virtual human, or avatar, created by Nvidia to showcase their new Geforce graphics card.

The Dawn demo draws detailed graphics at high speeds.

In the past, each frame of her scenes would take a computer hours to draw, and the whole thing would be stuck together afterwards to make the movie.

If you wanted to draw each frame faster, in real-time, you would have to go for a less detailed, more cartoon-like character, like Lara Croft.

The Dawn demo takes the best of both worlds, with very detailed graphics being calculated and drawn at very high speed as the frames are needed.

"If we look at how games are done currently, you can see that although they have the right skin colour and the faces have eyes and noses and mouths, they lack the imperfections that make them look realistic," said Nvidia's Andrew Humbar.

"Typically, the closer you zoom into a real-time character, the worse the image quality gets. But the closer you get to Dawn, the better she gets.

"That's when you really start to see the detail. So if you look at her skin, she has moles; if you look at her eyes, she has a retina, she has a cornea and blood vessels in the eye."

150,000 triangles

The Geforce chip is made of copper instead of aluminium, which means it can run faster, at 500 MHz or thereabouts. It gets quite warm, which is why it comes complete with its own cooling system.

"If we can produce avatars that convincingly behave like people, that look like people, there's no reason why people won't accept avatars" Patrick Olivier, Lexicle

It is this extra speed that allows the chip to handle a similar quality of image to those in recent animated movies.

"Dawn is made up of over 150,000 triangles. These triangles form a mesh, and this mesh is controlled to create the very realistic movement that we see in this particular demo," said Mr Humbar.

"The smaller the triangles, and the more triangles there are, the more control we can have over that particular image.

"If you look at Dawn's face and you can see all the different emotions, they're all being created by being able to manipulate and change the attributes of those triangles."

Because the images can now be drawn almost instantly, as they are needed on screen, it means avatars can now both look very detailed, and react to you in real-time.

Human imperfections

Looking good is important, but if an avatar is to be totally life-like, it will have to sound good too. How do you give a computer a human voice? Do you pre-record every single English word, with every single inflection?

It was a problem faced by Jonathan Jowitt, when he invented the news reading avatar Ananova.

"Most avatar systems that are on the market today use a process of converting written text into audio," said Mr Jowitt.

Is Ananova the news reader of the future ?

"In previous times, a text-to-speech engine would look at how are the words are constructed, and try to reassemble that in an audio domain, using short phonetic sounds.

"Things have moved on, so that engines these days know combinations of letters and word clusters. Our new text-to-speech engine apparently has the word 'the' in 700 times, which is impossible to believe, but some of the pronunciations of 'the' are very short."

Just as the key to looking human is the imperfections, it is important that the avatar does not sound too perfect either.

"We've actually tried to help the suspension of disbelief, by adding in the guttural and mouth sounds that you wouldn't get out of a text-to-speech engine.

"Our version of the text-to-speech engine has breathing, it has little throat clears in there, and little tongue tick sounds that are in between sentences. That certainly adds a layer of human element to it that wouldn't be in the engine in the first place," said Mr Jowitt.

Human or zombie?

The next challenge is to ensure the avatar behaves like a human.

Cara, the online customer service avatar developed by Lexicle, is designed to put customers at ease by displaying a more subtle human trait - body language.

Cara puts a human face on financial transactions

"If you think about it, what do people do when they speak?" asked Patrick Olivier, a computer linguist with Lexicle.

"They move their hands, they gesture, their facial expression changes, their lips move; and you tend to beat your hands, nod your head on words that you're trying to emphasise in that sentence, on words that are important in that sentence.

"It's critical to be able to compute those gestures automatically if you want your avatar to be convincing. Although you can get the graphical stuff right, you don't know what to make that avatar do unless it's got a brain driving it behind it."

Avatars seem to be getting ever more lifelike, but even here, there is a danger. As your avatar approaches reality it could fall into what is called the Zombie Zone.

A Japanese researcher has investigated how people react to avatars as they become more and more real, or naturalistic.

"He discovered that people's reactions to them improved quite nicely until they were almost naturalistic, and then there was a really sudden drop," said Ruth Aylett, professor of intelligent virtual environments at the University of Salford.

"The explanation for this was that you develop expectations about the character, that it is actually human, that it is actually real.

"Then suddenly those expectations are violated by something that slightly wrong in the voice, or the face, or in the way it moves, and it gives you a horrible feeling that is not real.

"And that's why we call it the zombie zone informally, because it's almost as if it was a zombie. It's the living dead."

Although there is a fair way to go before they become totally lifelike, avatars are even now working their way into every corner of our lives.

You can already create an avatar of yourself to attach to e-mails, or send to your mobile phone, or just to do the things you would never do.

"If we can produce avatars that convincingly behave like people, that look like people, and hopefully can actually perform tasks better than people, there's no reason why people won't accept avatars," said Lexicle's Patrick Olivier.

"They're going to make everyone's life easier."

http://news.bbc.co.uk/2/hi/technology/3096415.stm

ssander9 · 08-01-2003, 03:54 AM

Thanks for the update on avatars. Dawn Fairy looks impressive.
BTW, she has more than the eyes, blood vessels and retnas going for her. That outfit is awesome lol

07-27-2003, 07:13 PM	#1 (permalink)
Tandem Insane Location: Brisbane, Australia	Virtual humans edge closer By Spencer Kelly BBC For years, one of the main goals in computer graphics has been to recreate a totally convincing human being on screen, something that looks and acts so life-like that it is indistinguishable from a real person. But human beings are hard to recreate, because along with all those tiny details in the way we look, there is the way we move, our expressions. Now there is a new type of computer graphics card which is bringing virtual humans a lot closer. Dawn Fairy is a virtual human, or avatar, created by Nvidia to showcase their new Geforce graphics card. The Dawn demo draws detailed graphics at high speeds. In the past, each frame of her scenes would take a computer hours to draw, and the whole thing would be stuck together afterwards to make the movie. If you wanted to draw each frame faster, in real-time, you would have to go for a less detailed, more cartoon-like character, like Lara Croft. The Dawn demo takes the best of both worlds, with very detailed graphics being calculated and drawn at very high speed as the frames are needed. "If we look at how games are done currently, you can see that although they have the right skin colour and the faces have eyes and noses and mouths, they lack the imperfections that make them look realistic," said Nvidia's Andrew Humbar. "Typically, the closer you zoom into a real-time character, the worse the image quality gets. But the closer you get to Dawn, the better she gets. "That's when you really start to see the detail. So if you look at her skin, she has moles; if you look at her eyes, she has a retina, she has a cornea and blood vessels in the eye." 150,000 triangles The Geforce chip is made of copper instead of aluminium, which means it can run faster, at 500 MHz or thereabouts. It gets quite warm, which is why it comes complete with its own cooling system. "If we can produce avatars that convincingly behave like people, that look like people, there's no reason why people won't accept avatars" Patrick Olivier, Lexicle It is this extra speed that allows the chip to handle a similar quality of image to those in recent animated movies. "Dawn is made up of over 150,000 triangles. These triangles form a mesh, and this mesh is controlled to create the very realistic movement that we see in this particular demo," said Mr Humbar. "The smaller the triangles, and the more triangles there are, the more control we can have over that particular image. "If you look at Dawn's face and you can see all the different emotions, they're all being created by being able to manipulate and change the attributes of those triangles." Because the images can now be drawn almost instantly, as they are needed on screen, it means avatars can now both look very detailed, and react to you in real-time. Human imperfections Looking good is important, but if an avatar is to be totally life-like, it will have to sound good too. How do you give a computer a human voice? Do you pre-record every single English word, with every single inflection? It was a problem faced by Jonathan Jowitt, when he invented the news reading avatar Ananova. "Most avatar systems that are on the market today use a process of converting written text into audio," said Mr Jowitt. Is Ananova the news reader of the future ? "In previous times, a text-to-speech engine would look at how are the words are constructed, and try to reassemble that in an audio domain, using short phonetic sounds. "Things have moved on, so that engines these days know combinations of letters and word clusters. Our new text-to-speech engine apparently has the word 'the' in 700 times, which is impossible to believe, but some of the pronunciations of 'the' are very short." Just as the key to looking human is the imperfections, it is important that the avatar does not sound too perfect either. "We've actually tried to help the suspension of disbelief, by adding in the guttural and mouth sounds that you wouldn't get out of a text-to-speech engine. "Our version of the text-to-speech engine has breathing, it has little throat clears in there, and little tongue tick sounds that are in between sentences. That certainly adds a layer of human element to it that wouldn't be in the engine in the first place," said Mr Jowitt. Human or zombie? The next challenge is to ensure the avatar behaves like a human. Cara, the online customer service avatar developed by Lexicle, is designed to put customers at ease by displaying a more subtle human trait - body language. Cara puts a human face on financial transactions "If you think about it, what do people do when they speak?" asked Patrick Olivier, a computer linguist with Lexicle. "They move their hands, they gesture, their facial expression changes, their lips move; and you tend to beat your hands, nod your head on words that you're trying to emphasise in that sentence, on words that are important in that sentence. "It's critical to be able to compute those gestures automatically if you want your avatar to be convincing. Although you can get the graphical stuff right, you don't know what to make that avatar do unless it's got a brain driving it behind it." Avatars seem to be getting ever more lifelike, but even here, there is a danger. As your avatar approaches reality it could fall into what is called the Zombie Zone. A Japanese researcher has investigated how people react to avatars as they become more and more real, or naturalistic. "He discovered that people's reactions to them improved quite nicely until they were almost naturalistic, and then there was a really sudden drop," said Ruth Aylett, professor of intelligent virtual environments at the University of Salford. "The explanation for this was that you develop expectations about the character, that it is actually human, that it is actually real. "Then suddenly those expectations are violated by something that slightly wrong in the voice, or the face, or in the way it moves, and it gives you a horrible feeling that is not real. "And that's why we call it the zombie zone informally, because it's almost as if it was a zombie. It's the living dead." Although there is a fair way to go before they become totally lifelike, avatars are even now working their way into every corner of our lives. You can already create an avatar of yourself to attach to e-mails, or send to your mobile phone, or just to do the things you would never do. "If we can produce avatars that convincingly behave like people, that look like people, and hopefully can actually perform tasks better than people, there's no reason why people won't accept avatars," said Lexicle's Patrick Olivier. "They're going to make everyone's life easier." http://news.bbc.co.uk/2/hi/technology/3096415.stm

08-01-2003, 03:54 AM	#2 (permalink)
ssander9 Amplitude Modulator Location: US	Thanks for the update on avatars. Dawn Fairy looks impressive. BTW, she has more than the eyes, blood vessels and retnas going for her. That outfit is awesome lol __________________ I never said you had to.