Microsoft has filed a patent describing AI that could alter a speaker’s words, face and mouth during a live video. The proposed uses range from translation and removing filler words to filtering speech a viewer finds offensive or inappropriate, putting a consequential idea on paper without showing that it has become a product.
Microsoft AI Watch analysis
What happened
Futurism reports that the patent describes AI agents editing live video and audio, with adjustments made around replaced or removed words to smooth the transitions. The patent suggests viewers could tailor what they hear, while the original audio and video might be recorded as “ground truth”.
Its proposed examples include translation, removing “um” or “ah”, and changing mispronounced, misspoken or offensive words. It also describes removing colloquial language, including “y’all”. Those are possibilities set out in a patent, not evidence that Microsoft has built or plans to launch the system.
Why it matters
Live translation could make broadcasts more accessible. Personalised edits could also make a live speaker appear to say something different, or quietly strip away the dialect and mannerisms that make their speech theirs. A stored original might help preserve a reference, but viewers would still need to know when the version in front of them had been changed.
Our read
This is a patent filing, not a product announcement, so there is no need to start checking your television for a settings menu. But the proposal makes a useful distinction hard to ignore: translating someone’s words is not the same as editing their speech to suit an individual viewer. If a tool like this ever reaches audiences, clear labelling and access to the original would matter as much as the polish of the edit.
What to watch
- Whether Microsoft develops or announces a product based on the patent.
- Whether any future version clearly labels edits and preserves access to the original.
- How the company would distinguish translation from changes to a speaker’s wording or appearance.
Discussion spark: If a live-video service edits a speaker’s words or face for each viewer, should it be treated as translation, censorship, or a new version of the event?
Sources and evidence
- Microsoft Patents an AI That Can Literally Put Words in People’s Mouths in Real Time – Futurism (1 November 2026, 17:09 UTC)
not affiliated with or endorsed by Microsoft