Experimental Tech Restores Roger Ebert's Voice

Scottish company reconstructs Ebert's voice from hours of movie commentary.

March 2, 2010, 10:57 AM

March 2, 2010— -- After a more than three-year journey to find his voice, new technology is helping Roger Ebert sound like himself again.

On "The Oprah Winfrey Show" today, the celebrated film critic, who lost his voice after complications from thyroid cancer, demonstrated new voice cloning technology that will let him communicate in his old voice.

Using hours of archived recordings of Ebert's "At the Movies" commentary, the Scotland-based company CereProc reconstructed the sound of the critic's voice. Ebert himself found the company while searching the Internet and reached out to them last year.

CereProc, an advanced voice synthesis company in Edinburgh, creates customized text-to-speech software. Instead of producing flat computerized voices, the company says its voices include realistic, animated and emotional dimensions.

After demonstrating the technology on Oprah, Ebert gave his new voice the thumbs up.

"You'll know it's a computer, but one that sounds like me," he said. "It still needs improvement but at least it sounds like me. In first grade they said I talked too much, and now I still can."

Moved by the sound of a voice she said she had not heard since July 1, 2006, Ebert's wife Chaz said, "I actually think it's incredible. It's incredible that that's your voice."

Ebert used a prototype for Winfrey's show and expects to receive a final version of the technology in the next few months.

Ebert said the new voice is "uncanny, a good feeling."

Ebert was diagnosed with thyroid cancer in 2002, but at the time neither he nor his doctors expected him to lose his voice.

The cancer spread to his throat, and attempts to treat the spreading cancer with radiation led to more complications. Several subsequent surgeries meant to restore his voice failed.

In the effort to communicate again through speech, his new "voice" went through several incarnations before settling on the CereProc technology.

In a piece published last August on his journal on the Chicago Sun-Times' Web site, Ebert wrote about his experiments with text-to-speech technology. On his Mac, he said he found a speech program that reads aloud messages typed on his laptop.

Though he called it a "godsend," he said it still left much to be desired. He tested out different voices and, after using a British accented voice called "Lawrence" for a year, he ultimately settled on "Alex."

"Amazingly Alex, is the first voice I have yet heard that understands and reflects question marks and exclamation points!?! This prevents every sentence from being spoken in a monotone," Ebert wrote.

Ebert Found CereProc on His Own

Still, Ebert said he wished for his own voice again.

"On those occasions I've appeared in public or on TV with a computer voice, I nevertheless sound like Robby the Robot. Eloquence and intonation are impossible. I dream of hearing a voice something like my own," he wrote. "We put men on the Moon, people like to say about such desires: Why can't I have a voice of my own?"

In that same piece, he wrote about reaching out to CereProc.

"I have my fingers crossed. I have launched an e-mail to Edinburgh with my appeal. I can see my own voice hosting online or telecast video essays," he wrote. "I am greatly cheered. I will keep you informed."

Matthew Aylett, CereProc's chief technical officer, said the company started working on the voice Ebert has affectionately called "Roger, Jr." soon after receiving his request.

"We typically use synthesis in all sorts of environments, but this is the first time we've used it in a way that really helps someone who needs it in this way," he said.

Usually, the company's text-to-speech technology is used by companies that need a synthesized voice for call centers or in-car communication, he said. Academics also work with CereProc on voices for more human-sounding, cutting-edge robots. But Aylett said that as the technology advances, the applications are endless.

In all cases, Aylett said, CereProc engineers and speech experts work to "infuse character in our voices."

In addition to creating regional-sounding voices, the company produces voices with variation and emotions. Though CereProc can't yet recreate the full range of emotions,it's possible to construct voices that sound more snippy, cheerful or sad, he said.

The work on "Roger, Jr." builds on efforts to clone speech.

"When you clone a voice, what you're trying to do is represent a character and get that personality out," he said. Using recorded audio of George W. Bush, for example, the company recreated the voice of the former president.

The prototype voice that Ebert demonstrated for Winfrey was constructed from about three hours of the critic's commentary on movies such as "Citizen Kane" and "Valley of the Dolls."

After transcribing the commentary, Aylett said the company broke down the words into the roughly 45 speech sounds that comprise the English language. Those sounds are also affected by Ebert's intonation patterns and the ways he expresses feeling in a sentence, he added.

The task was made more difficult because Ebert's commentary was recorded over different years, in different audio environments and with different equipment. Aylett said the transcription process was complicated by the fact that Ebert frequently spoke off the cuff.

Still, he said the company continues to add data to build Ebert's voice and expects to have a completed voice in the next few months.

"The voice communicates an awful lot about you," he said. "Your history, where you're from, what sort of person you are in many ways. It's also part of our soul. To create that, for example, for people who have lost it is a really exciting thing to do."

ABC News' Lauren Cox and Emily Friedman contributed to this story.