Microsoft has a patent application for software that would edit speech in a live video stream and then reshape the speaker’s lips to hide the join. It is paperwork only: the company has said nothing about turning it into a product.
The application, US 2026/0341309, is titled “Near Real-Time AI Enhancement of Video or AR Presentations of Live Events” and was published by the US patent office on 29 October. AI agents would find words in a live feed, replace or delete them, and adjust the speaker’s face and audio around the change so it looks smooth. One report puts the delay at roughly 1 to 10 seconds.
The patent lists several reasons to do it. The simplest are filler words such as “um”, stutters and mispronounced words. It also names slang, local phrases and offensive or inappropriate words, with replacements possibly pulled from an internet search. A separate use is live translation with lip-sync, so a viewer sees a broadcast in their own language. Substitutions could be tuned per viewer, using details such as their age, place and language.
The per-viewer part is what worries critics: two people could watch the same live event and see different words coming out of the speaker’s mouth. The filing does nod to the risk, saying a copy of the unaltered feed can be stored as a “ground truth” reference. It names lectures, presentations, TV shows and live sports as places the system could run.
Companies file patents for ideas they never ship, and a published application does not mean the patent has been granted. No product name or date is attached to this one.
