Imagine watching someone speak during a livestream. Their voice sounds normal, their facial expressions look natural, and their lips match every word.
Except some of those words might not be what they actually said.
That is the possibility raised by a newly published Microsoft patent application describing artificial intelligence capable of modifying speech and facial movements in near real time.
The proposed system would analyze audio and video, identify words that could be changed, and generate replacement speech. It would then adjust the speaker’s mouth and surrounding facial movements to make the new words appear natural.
The technology could have practical uses.
For example, imagine attending an online lecture in English but watching the speaker appear to communicate fluently in French or Arabic, with their lip movements matching the translated speech.
It could also remove verbal mistakes, filler words and awkward pauses from live presentations.
But there’s another side to the technology.
If AI can make someone appear to say words they never actually spoke, how will audiences know whether a video is authentic?
Deepfakes have already raised concerns about misinformation, impersonation and manipulated online content. Technology that could make alterations during broadcasts would introduce another challenge.
Microsoft’s application, published on October 29, describes how such a system might work, but it is not an announced product.
There is no confirmation that Microsoft has developed or deployed a working commercial version, and patent applications do not guarantee that a technology will reach the market.
Still, the concept highlights an increasingly important question about digital media.
For years, video recordings have been treated as strong evidence of what someone said or did. As AI editing becomes more sophisticated, that assumption may become harder to maintain.
The technology could make communication more accessible. It could also make verifying what is real more important than ever.
