Software restores a critical sound
Company re-creates Roger Ebert’s voice
CHICAGO – Nearly four years after a battle with thyroid cancer robbed him of the ability to speak, iconic film critic Roger Ebert sounded like his former self Friday during a taping of “The Oprah Winfrey Show,” the show’s producer said.
It was no medical miracle, but rather a demonstration of new software using audio recordings of Ebert to create a synthetic voice that sounds like his own.
CereProc, a company based in Edinburgh, Scotland, created the voice for him using mostly audio of Ebert’s DVD commentaries on “Citizen Kane” and “Casablanca.”
The company’s technology allows Ebert to sound more natural than other “text to speech” software – even allowing for a range of emotions.
“Roger has many years of experience in broadcasting,” said Matthew Aylett, chief technical officer for CereProc. “Obviously we couldn’t record him but he did have a lot of audio material we could use to build his voice.”
The company has used the technology – which turns text typed by the user into sound – to build voices of other famous people, including former President George W. Bush on a satirical Web site.
But this is the first time the company has produced a synthetic voice that sounds like the old voice of the person using it, Aylett said.
Ebert could not be reached for comment Friday, but in a blog post last summer, he described his frustrations with trying to communicate.
After his second surgery, Ebert learned he would no longer be able to speak and started writing notes, he wrote in an August 2009 blog entry. But he found that took too long to keep up with normal conversation.
“There is a point when a zinger is perfectly timed, and a point when it is pointless,” Ebert said.
At the time, the Chicago Sun-Times film critic said he had been experimenting with synthetic voice software made by other companies. He tried a voice named Lawrence, which had a British accent, and more recently a voice named Alex, which had an American accent that sounded more natural because it recognized punctuation marks, Ebert wrote on his blog.
But he still hoped to sound more like himself, particularly at public appearances, he said.
“On those occasions I’ve appeared in public or on TV with a computer voice, I nevertheless sound like Robby the Robot,” he wrote on his blog. “Eloquence and intonation are impossible. I dream of hearing a voice something like my own.”
The taping of Ebert on Winfrey’s program, which will air Tuesday, includes him giving the talk show host his 2010 Oscar picks and allowing her cameras to follow him for a day, the show’s producer said.
While Ebert’s new voice sounds like his own, it occasionally makes errors, Aylett said. In particular, the software has difficulty pronouncing unusual proper names and sometimes fails to make intonation sound natural, he said.
“It sounds like him,” he said. “But it will sound better as we add more audio information to it. The more data we have, the smoother and the more accurate the voice will become.”
On its Web site, CereProc says it provides voices that “sound real” and “have character.” To build Ebert’s voice, the company is using between three and five hours of his voice recordings and cutting them into numerous small units of sound. The software also allows for users to insert emotions, from anger to happiness.
The latter would seem to best describe Ebert’s reaction to the software’s possibilities. Last summer, after discovering CereProc while surfing the Internet, Ebert said he had big plans, including using his own voice to host online or telecast video essays, he wrote on his blog.
“I am greatly cheered,” he wrote.
Even while his new voice is being fine-tuned, the software offers Ebert a chance to get back a semblance of what he lost, Aylett said.
“Your voice is very much a part of who you are, your history and identity,” he said. “We’re giving Roger back his power of speech, to a certain extent.”