AI has become adept at recreating, altering, and restoring human speech. But as the replicas become indistinguishable from the real, fears about the tech are growing.
Alex Serdiuk has a unique understanding of both the opportunities and threats.
As the co-founder and CEO of AI startup Respeecher, Serdiuk has won an Emmy for creating a deepfake Richard Nixon, developed voice clones for speech disabilities, and de-aged Mark Hamill’s vocal cords for The Mandalorian.
Greetings, humanoids
Subscribe to our newsletter now for a weekly recap of our favorite AI stories in your inbox.
Yet Serdiuk has also seen synthetic media at its worst. The CEO and his company are based in Ukraine, which has been the target of deepfake disinformation.
In March, a manipulated video of Ukrainian President Volodymyr Zelensky circulated on social media. The clip showed a digitally-rendered Zelensky telling soldier to surrender to Russia.
The impact, however, was minimal.
“This deepfake was done so poorly — like many things that Russians do — so it wasn’t convincing,” Seriduk tells TNW.
“And our nation is smart. We have a belief in what’s going on in our government, and if someone says that our president gave up, most of the people would check those facts — especially because the deepfake was so bad.”
The Zelensky deepfake is so bad it won’t convince any Ukrainians to “lay down their arms.” But enough of these could spread doubt about the authenticity of real videos in the future https://t.co/DZ7IuYsPoT pic.twitter.com/MGNxvyw8GE
— Alec Luhn (@ASLuhn) March 17, 2022
Nonetheless, the clip demonstrated synthetic media’s potential to make us question what we see — and hear.
Indeed, fake sounds can be more convincing than fake sights.
Editing reality
Seriduk believes synthetic voices can avoid the uncanny valley more smoothly than artificial visuals.
He adds that this realism can benefit society. Respeecher, for instance, has developed voice replacement tech for people who have undergone a laryngectomy.
In trials, the system created a natural-sounding voice while preserving the user’s articulation.
Sonantic, an AI startup, produced another powerful example.
In 2021, the company recreated Val Kilmer‘s voice after throat cancer treatment let the actor unable to speak clearly.