First there was the Turing Test, a benchmark for computer intelligence via conversation.
Now perhaps, there is the Ebert Test, a way to see if a synthesized voice can deliver humor with the timing to make an audience laugh.
Last Friday Roger Ebert, the movie critic who lost his ability to speak several years ago after his lower jaw was removed, used a computer-created voice to tell a joke to an audience of over 1,000 people at the TED conference in Long Beach, Calif.
“A guy goes into into a psychiatrist. The psychiatrist says, ‘You’re crazy.’ The guy says, ‘I want a second opinion.’ The psychiatrist says, ‘All right, you’re ugly, too.'”
Sure, it was the corny standby stand-up joke. But the punchline sent the audience erupting into laughter, which delighted Mr. Ebert. He proposed the Ebert Test as a way to gauge the humanness of a synthesized voice.
“If the computer can successfully tell a joke, and do the timing and delivery, as well as Henny Youngman, then that’s the voice I want,” Mr. Ebert said.
Through a combination of his Macintosh laptop and readings of a prepared script by friends and his wife, Mr. Ebert told the audience of his multiyear struggle since he lost his voice in 2006. He cannot speak because his mouth can’t close to trap the air that is needed to produce oral sounds.
“All my life I was a motor-mouth. Now I’ve spoken my last words, and I don’t even remember, for sure, what they were,” he said via his friend John Hunter.
Mr. Ebert is working with a Scottish company called Cereproc, founded in 2005, to recreate his old voice using recorded movie commentary. The voice, called Roger Jr., wasn’t yet ready for the TED stage.
“I produced the full speech using our voice,” Matthew Aylett, the chief technology officer of Cereproc, said in an interview. “I wasn’t completely happy with it myself.”
He said the voice the company had created for Mr. Ebert was much more casual in tone. “The big issue with public speaking is that there is a rhetorical way of speaking.”
But Mr. Ebert noted, “Because of the digital revolution, I have a voice, and I do not need to scream.”
He opened his speech with a clip from the movie “2001: A Space Odyssey,” featuring Hal 9000, which was
supposedly created in Mr. Ebert’s hometown of Urbana, Ill. He said his wife, Chaz,
loved a voice named Lawrence, who had a British accent.
Mr. Ebert said he was highly satisfied with a Macintosh voice named Alex, which he used on stage, because it could tell the difference between a period, question mark and exclamation point.
But last year, on an emotional appearance on the “Oprah Winfrey Show,” Mr. Ebert unveiled his new voice from Cereproc, just months after he started working with the company. Since then, Mr. Aylett says they have set ambitions to create a broadcast-quality voice, which is challenging because it is both articulate yet engaging.
The ability to deliver that depends on Mr. Ebert’s recorded commentary. In movies that he is less familiar with, he is more formal, but in most of the recordings, he is very relaxed and casual.
“No one has really has done a synthesis from found material,” Mr. Aylett said, noting that most voice creations had involved actors who could be given hours of materials to be read aloud. “In a way, he has been able to help us move this forward.”
Synthesized voices have become popular digital accessories. Among the latest offerings: Yoda and Darth Vader can both be downloaded as voices for TomTom’s GPS navigation devices.
Cereproc has created a battery of voices, including ones for Barack Obama, Arnold
Schwarzenegger, and George W. Bush (which can be instructed to say anything you type in a box).
Other researchers are making inroads into creating human-like voices. I.B.M. has patented a technology that has been programmed to be less robotic, by including verbal tics like “ums” and sighs, pausing for effect or coughing to attract attention.
But, as Mr. Ebert’s presentation at TED indicated, humor is still one of the challenges for computerized speech.
“What it shows when you tell a joke is that you have mastery of your delivery,” said Mr. Aylett, who added that even people are not always able to tell jokes well. “There is a skill and intelligence that people have when they speak different ways that is really incredible. It’s actually hard to pin down what that really is.”
The first step may be to have mastery of humor, but not in real time. “To take a joke and craft it, you can change the timing,” Mr. Aylett said. One issue is that although humans can type around 80 words per minute, they speak at about 130 to 180 words per minute.
A few years ago, Cereproc worked with the Ford Motor Company to create a voice to tell jokes in the car. “We had to have them give us the jokes. I didn’t want my engineers to make them up, because people’s sense of humor are different,” he said.
While humor — both delivery and creation — is considered by philosophers and computer scientists as a core element in being human, the difficulty of achieving that through research is not always widely appreciated. Last year, Senator John McCain listed a $712,000 Northwestern University research grant for computer-generated humor as No. 36 of 100 wasteful stimulus spending projects in a report.