BBC Radio Deepfake

BBC Radio Hosts Use Deepfake AI to Swap Voices


BBC Radio 1 Breakfast hosts Greg James and Matt Edmondson amused and worried listeners this month when they used deepfake technology to speak in each other’s voices. James and Edmondson poked fun at each other with the swapped voices in a bit inspired by the BBC-produced TV thriller The Capture using voice clone models built with synthetic speech AI firm Respeecher’s help.

Capture Clones

The Capture’s recent second season dives into the potential for deepfakes and synthetic media and how unethical use can potentially ruin people’s lives. The show tells the story of a British security minister dealing with the fallout of a deepfake video showing him taking the opposite of his actual political position. The radio show hosts went for a lighter game of teasing and using the other’s voice to offer to buy everyone lunch or make fun of their pet.

“From the very first briefing for The Capture Series 2 we knew we had to get some of the hypothetical tech used in the series and show it was possible today in a LIVE broadcast,” senior creative at BBC Creative Michael Tsim wrote in a LinkedIn post. “We didn’t know how, or whether it was even possible. But after literally months of negotiations, contracts and conversations with some incredibly clever guys in Ukraine (Respeecher!) , we actually managed to pull it off. A live, completely real deepfake stunt with the brilliant Greg James and Matt Edmonson on Radio 1.”

Deepfake Doubts

While many listeners enjoyed the segment, some were both impressed and concerned, with responses on Twitter and TikTok expressing anxiety and even fear about the potential misuse of synthetic speech and deepfakes. “This was brilliant, but a bit scary tbf,” wrote one TikTok viewer, while another tweeted, “Amazing and terrifying!” Others had much stronger words against the whole idea of creating deepfake technology.

‘The Capture’ Crystallizes Deepfake Fears on TV

Synthetic Steve Jobs Interviewed by Deepfake Joe Rogan for AI-Powered Podcast Scripted by GPT-3