For longtime listeners of NPR’s Planet Money, there are few voices as recognizable and iconic as that of Robert Smith, one of the show’s former hosts. But even the most experienced and discerning ears may struggle to distinguish between the journalist and his vocal clone.
Accuracy is inspiring or terrifying, depending on your perspective.
Either way, it’s a credit to the technology behind WellSaid Labs.
The Seattle startup, which grew out of Seattle’s Allen Institute for Artificial Intelligence in 2019, has agreed to create Synthetic Robert for a three-part Planet Money series in which co-hosts Jeff Guo and Kenny Malone used artificial intelligence to produce anything from research to interview. questions to the episode script and even a radio drama.
The result was an AI-powered show, which debuted Friday night, with Smith’s synthetic voice and Malone’s real voice co-hosts. I won’t spoil the ending, but the whole thing is surprising, enlightening, and a little scary.
Cloning the human voice is becoming more common in the industry, but it’s not the norm for WellSaid Labs. The startup focuses on custom synthetic voices tailored to the needs of its customers, not exactly replicating real human voices.
We followed WellSaid Labs to find that this exercise in precise voice replication represents either a new direction for the company or just a one-time experiment for the popular NPR show.
In particular, the company has placed elaborate conditions on Planet Money’s use of the technology.
As explained by Guo and Malone in the second episode of the series, WellSaid Labs requested explicit permission from Smith to create the vocal clone. The company also monitored every word Planet Money had said to Synthetic Robert, under threat of ending the entire exercise if it was used for anything that didn’t align with the show’s values.
And perhaps the biggest term and condition of all, as soon as we were done with this project, Synthetic Robert would be shut down, Guo explains in the second episode. It could narrate our AI-generated episode. And then it will be functionally destroyed, never to be used again.
While our efforts pale in comparison to the epic Planey Money project, we tried a more modest version of this experiment on the GeekWire podcast a few weeks ago, using voice clones to read an AI-generated script.
At the time, WellSaid Labs declined our invitation to create vocal clones of my GeekWire colleague John Cook and myself. While we would have preferred to support a Seattle startup and leverage AI technology from our own backyard, we instead used technology from New York-based startup ElevenLabs.
ElevenLabs offers DIY vocal cloning of real human voices, based on vocal samples. It requires the user to confirm that the human whose voice is being cloned has given permission, but that’s basically a checkbox, not a strict safeguard. The ElevenLabs vocal clone was perfect for John, but a little out of place for me, even after extensive fine-tuning.
One big difference: We were able to create ElevenLabs voice clones for the GeekWire podcast in minutes. Planet Money had to wait a couple of weeks for WellSaid to create Synthetic Robert, according to the show.
It was worth the wait.
Rhyan Johnson of WellSaid Labs joins Guo and Malone in the second episode to play clips illustrating the evolution of the AI-generated version of Smith’s voice. What starts out as a jumbled mess turns into a nearly perfect replica, making it hard to tell the difference between the silicon and carbon versions of the radio host.
Catch up with Planet Money’s three-part AI series that starts here.
#synthetic #NPR #host #startup #WellSaid #Labs #creates #amazing #vocal #clone #Planet #Money