Chinese state media has TV news anchors that can broadcast 24/7 with computer-generated faces
Xinhua’s latest news anchor looks like a person. He sounds (kinda) like a person.
But he can also work 24 hours a day, seven days a week… and read multiple stories on multiple channels at the same time.
The Chinese state-run news network unveiled two “AI synthesized anchors” yesterday. They’re based on real people, but their faces and voices are computer-generated -- allowing them to say and do things that their real models never did.
“My appearance and voice are based on Xinhua anchor Qiu Hao, but I never need to rest,” the AI says in its debut video.
The software, developed by search engine Sogou, learns a human anchor’s voice and face by watching his newscasts. It uses speech synthesis to make it “read” text inputs, and generates lip movements and facial expressions to match the words in real time.
The result is impressive… and a little creepy. It looks good enough, but there’s something odd about the way their lips move. And the voice sounds like, well, any other computer voice -- it has the same stilted delivery as Siri or Alexa.
Chinese netizens aren’t blown away either. There aren’t many reactions on social media, but most find it a bit disturbing.
“Pretty impressive, but it still feels weird,” one user says on Weibo.
Another says, “Can we not use this kind of voice? It’s really creepy.”
When asked by Chinese media if their new anchors will take jobs away from humans, Sogou’s CEO said, “In terms of perception of sound and image, machines may be able to be as good as human, but they have limited abilities when it comes to cognition, meaning reasoning, thinking and logics based on language.” Which suggests that reading the news is a job that machines may take over -- but the things they read will still need to come from humans.
You’ve probably heard about “AI news anchors” before. A Japanese robot named Erica was set to be a news anchor “sometime in 2018” (it hasn’t happened yet), and Chinese TV stations have also tried using Microsoft’s voice assistant XiaoIce to broadcast weather and letting a robot host a show. But Xinhua’s AI avatar could be the first time it has a face based on a real human.
Sogou first announced the “virtual anchor” technology in July this year, when the company said it could also be integrated with Sogou’s own voice assistant to generate more characters to be used in entertainment, education, medical and even in legal situations.
It’s a new use, but the technology is not entirely new. It was famously demonstrated on President Obama, where his lips are synced to words he never said. Similar tools have also been more accessible to people, allowing deepfake videos to flow around, causing ethical concerns.
Xinhua claims that the AI anchors will be seen across Xinhua’s platforms, including on its app, website and mini-program in WeChat.
Their potential spread left some to lament one thing that might be lost in the future: “We won’t be able to see TV news bloopers anymore.”