AI voice-clone of legendary sportscaster Al Michaels will narrate personalized daily video recaps during 2024 Olympics

[This story from Vanity Fair is being cited in much of the coverage of an announcement by the NBC television network that a simulated version of legendary broadcaster Al Michaels’ voice will be used to create personalized 10-minute daily highlight videos during the 2024 Summer Olympics. The NBC press release about this latest milestone in artificial intelligence and presence notes that

“’Your Daily Olympic Recap on Peacock [NBC’s streaming platform]’ will be featured in Peacock’s homepage and Olympics hub and available at the individual profile level, so up to six users on one account can receive their own recaps. The majority of users can also be welcomed by their first name for an even more personalized experience. Users can opt into receiving push and in-app notifications to remind them to watch their recaps.”

–Matthew]

“It Was Astonishing”: How NBC Convinced Al Michaels to Embrace His AI Voice for Olympics Coverage

The network will use an artificial clone of the legendary broadcaster’s voice to narrate its daily recaps of the summer event. “It was not only close,” he says of the technology, “it was almost 2% off perfect.”

By Tom Kludt
June 26, 2024

Few voices in American life are more recognizable than the one belonging to Al Michaels—play-by-play announcer for nearly a dozen Super Bowls and the source of perhaps the most famous line in sports history.

For generations of sports fans, Michaels has been a near-constant presence, providing the soundtrack of last-second field goals, ninth-inning walk-offs, and fourth-quarter buzzer-beaters. He was the voice of Monday Night Football for 20 years, then Sunday Night Football for 16. When the 1989 World Series was disrupted by an earthquake, Michaels’s voice was the one viewers heard just as the broadcast went static. And when a plucky United States hockey team pulled off an upset for the ages against the Soviet Union at the 1980 Olympics, Michaels channeled the prevailing sense of disbelief with a call as iconic as the game itself. (“Do you believe in miracles? Yes!”)

Michaels’s vocal stylings are typified by a breezy command of the English language, unique rhythms—like the emphasis he places on certain syllables of a player’s name (“A first down by Adam Thie-len”)—and a distinct but somewhat unplaceable regional dialect. The writer Drew Magary once described the Michaels cadence as “a ’50s Brooklyn accent that is now so obscure it counts as a dead tongue.”

So when NBC approached him with an idea to recreate his voice using artificial intelligence for its coverage of this summer’s Olympics, Michaels had a few reservations. That voice, after all, is both his livelihood and legacy.

“What would I sound like?” Michaels says. “Would I sound like a guy who just spews clichés? Would my voice be different?”

Michaels was “very skeptical” of the proposal—until he heard the AI for himself. “Frankly, it was astonishing. It was amazing,” he told me in a phone interview last weekend. “And it was a little bit frightening.” Michaels was left in awe of the nuance—the way it captured his intonations and verbal subtleties. “It was not only close, it was almost 2% off perfect,” he said. “I’m thinking, Whoa.”

With his concerns assuaged, Michaels gave his blessing to NBC, where he has held an emeritus role since 2022. His voice—or at least a highly convincing replica of it—will now be lent to a feature on the network’s streaming platform, Peacock, offering users daily recaps from the Summer Games in Paris tailored to their favorite events and narrated by the AI. NBC says it trained the AI to match Michaels’s delivery using his past appearances on the network. I heard it for myself in a demonstration provided by NBC last week, and sure enough, it sounded like the real Al Michaels.

“They were able to do exactly what I might—I shouldn’t say ‘exactly,’” Michaels said, catching himself before conceding too much to the bot. “It sounded like what I might say in certain situations,” he said.

Rick Cordella, the president of NBC Sports, said Michaels was the “perfect choice” for the feature. “Al deserves credit for leaning into this technology so enthusiastically,” he added.

The feature, called Your Daily Olympic Recap on Peacock, will pull from thousands of hours of live coverage from the Games in Paris using a large language model, or an LLM. The model analyzes subtitles and metadata to summarize clips from NBC’s Olympics coverage, and then adapts those summaries to fit Michaels’s signature style. The resulting text is then fed to a voice AI model—based on Michaels’s previous NBC appearances—that was trained to learn the unique pronunciations and intonations of certain words and phrases. In the end, this multilayered process will yield around 10 minutes’ worth of highlights for each user.

NBC says that there could be nearly 7 million personalized variants of the recaps, and that a team of human editors will review the content before it is released to users. (That layer of quality control will be especially important when it comes to the pronunciation of the athletes’ names.)

John Jelley, senior vice president of product and user experience at Peacock, said that the scale of the Olympics—the Paris Games will feature 32 sports and more than 300 medal events—made it the perfect place to deploy the technology. “It would be impossible to deliver a personalized experience with a legendary sportscaster to millions of fans without it,” Jelley said.

It also would have been impossible unless Michaels was fully on board. “This was born of curiosity because I’m a very curious person,” Michaels said. “I was approached about this project and didn’t really understand too much about it. Believe me, I am no techie by any stretch of the imagination, but I know that, ready or not, here comes artificial intelligence.”

Like many of us, Michaels has mixed feelings about the proliferation of artificial intelligence—and that goes beyond the realm of sports. For starters, he worries that it will stoke the flames of misinformation, which he called “the bane of our existence these days.”

“People are being thrown curveballs,” he said. “Can this be manipulated to the point where people are getting either catfished or gaslighted?”

He is also concerned about its implications on the workforce. A longtime Los Angeles resident, Michaels is close friends with a number of Hollywood screenwriters who have shared with him their anxieties about AI’s threat to their own livelihoods. “It could take jobs away from people, the writers who need to work,” he said.

But Michaels is also enthralled with AI’s potential for good, like its ability to unlock our understanding of diseases. “This is a pipe dream,” Michael said, “but if AI could someday take everything that’s ever been known and researched about cancer and somehow advance the curing of cancer—I mean, now that would be the all-time greatest thing that could happen.”

Michaels, 79, may not be a techie, but he is no Luddite either. He is fascinated by artificial intelligence, telling me he wants to learn as much about it as he can.

After NBC approached him about the Olympics project, Michaels’s curiosity was piqued and he decided to conduct a personal experiment. He went on ChatGPT and asked it to generate 10 plotlines for a modern-day adaptation of the 1950s-era sitcom Father Knows Best. Within seconds, the bot served up a variety of contemporary scenarios—one about the dad’s futile efforts to fix the Wi-Fi router, another about the dad enjoying unexpected viral fame after his kids teach him the ins and outs of social media. Michaels said he was “amazed and frightened at the same time.”

“I’m going, There has to be a man inside there. There’s a person inside there,” he marveled. “It knows what [the show] was. It takes the plot and advances it to 2024.”

He shared the results with his buddy Alec Berg, a longtime TV writer who served as cocreator of Barry, who Michaels says responded, “I may have to get into the plumbing business.”

Voice cloning through artificial intelligence has surged in recent months: It has been used by companies to improve efficiencies as well as by rank-and-file internet users to create so-called deepfakes. OpenAI, the company behind ChatGPT, revealed in March that it had created a tool that can recreate a person’s voice using just 15 seconds of recorded audio, but said it would not release it to the public due to concerns over potential misuse. Last year, the popular sports podcaster Bill Simmons revealed that his employer, Spotify, was developing AI to recreate its hosts’ voices for advertisements. Just last week, Universal Music Group inked a deal with an AI music-tech startup to help artists design their own voice clones.

[snip to end]


Comments


Leave a Reply

Your email address will not be published. Required fields are marked *

ISPR Presence News

Search ISPR Presence News:



Archives