Deepfake Queen Elizabeth II Will Deliver Alternative Speech and Dance on Channel 4
— Channel 4 (@Channel4) December 23, 2020
Queen Elizabeth II of Britain will follow her annual Christmas Day Message with an alternative on the publicly owned Channel 4, or at least that’s what it will look like at first glance. The Queen delivering Channel 4’s speech is actually a sophisticated deepfake, relying on artificial intelligence and a talented actress to mimic the monarch and deliver a decidedly out-of-character speech and dance routine.
The Queen’s standard Christmas speech will play on BBC 1 and ITV at 3 p.m. GMT. This year is already integrating new technology, with Alexa-enabled device owners able to ask the voice assistant to play the broadcast for the first time. Channel 4 has invited a wide range of notable series of public figures to give an Alternative Christmas Message afterward since 1993, ranging from disaster survivors and veterans to The Simpsons. To address concerns about technology spreading false information and misleading people, the station decided to demonstrate just how good the technology has become by recreating the Queen in pixels.
“This was a great project to be asked to direct. Deepfake is an interesting spin-off from the recent advances made in machine learning and AI and while it is a powerful new technique for image makers everywhere it is also a tool that can be used to misrepresent and deceive,” director William Bartlett said in a statement. “With Channel 4 we wanted to create a sequence that is hopefully entertaining enough that it will be seen by a lot of people and thereby spreads the very real message that images cannot always be trusted.”
Actress and impressionist Debra Stephenson performs the four-minute speech. Framestore, an Academy Award-winning video special effects firm, blended the audio with deepfake software to generate a video where the Queen warns of the dangers of deepfakes and misinformation. She also manages to throw in a genuinely funny joke about Canada before ending with a TikTok dance routine that is impressive even for a fake 94-year-old.
“For nearly seventy years, I have kept a tradition of speaking to you at Christmas,” the deepfake Queen says to begins her speech. “But on the BBC, I haven’t always been able to speak plainly and from the heart. So, I am grateful to Channel 4 for giving me the opportunity to say whatever I like, without anyone putting words in my mouth. If there is a theme to my message today, it is trust. Trust in what is genuine – and what is not.”
The visual fakery is impressive, but audio technology can make for even stranger deepfakes. For instance, researchers at MIT took the speech President Richard Nixon had prepared in case Apollo 11’s trip to the Moon failed and created a deepfake video of Nixon giving the speech. Both the audio and video are fictional, but it showcases the advances in both sides of the deepfake coin. The fake audio uses techniques similar to those Amazon used to give Alexa Samuel L. Jackson’s voice. Synthetic voice creation doesn’t require the resources of Amazon either. Startups like Resemble.ai can turn just three minutes of audio into a basic synthetic speech model. Even when there may be errors in the audio or video component, the use of both tends to smooth problems over. That said, if you see a nonagenarian on a screen dancing like an athletic 20-year-old, it’s probably not authentic.