THE ESSENTIALS IN 30 SECONDS

  • The filing covers near real-time AI enhancement of video or augmented reality presentations of live events, in which agents alter or remove spoken words and edit the speaker's face to fit.
  • Suggested uses include live translation, cutting filler words and stutters, and dropping words judged offensive; the original footage could be kept as a reference.
  • It was flagged by the tracker Patentlyze around October 30; Microsoft has not said whether it plans to build anything from it.

A Microsoft patent application that surfaced at the end of October sets out a system that could change what a person says during a live broadcast and then edit their face so the new words appear to come from their mouth. The document was first spotted by the patent tracker Patentlyze and was reported by Futurism on November 1, followed by outlets such as Digit and WION.

What the filing describes

The inventors present the idea as near real-time AI enhancement of video or augmented reality presentations of live events. According to Futurism's reading of the document, AI agents would detect certain words in a speaker's audio, then swap or delete them according to preset rules or individual viewer profiles. The surrounding sound and images would be adjusted to smooth the cut, and the speaker's mouth movements would be regenerated to match.

The listed uses are mostly practical: translating a talk into another language with matching lip movements, removing filler words, fixing stutters, and replacing words that are mumbled, mispronounced or flagged as offensive or inappropriate. Replacement terms could even be drawn from an internet search, Digit reports. Futurism says the filing names lectures, presentations, television formats such as awards, game and cooking shows, and live sport as possible settings. It also proposes recording the unaltered video and audio as a "ground truth" copy.

Why it draws attention

The concern raised in coverage is authenticity. If edits are personalised, different viewers could see different versions of the same event without knowing what was actually said, and Digit notes that regional expressions and accents could be smoothed away. The filing's archive of original footage is described as a safeguard, but it does not specify who would be able to consult it. Futurism also points out that dubbing and post-production editing already exist; the novelty here is doing it live and per viewer.

A patent is not a product

Companies routinely patent ideas they never ship, and Microsoft has not announced any plan to use this technology in a product or service. The coverage refers to a publication numbered US 2026/0341309 A1 on the US Patent and Trademark Office site, but its filing date and examination status were not given in the reports reviewed. Microsoft did not immediately respond to a request for comment from Futurism.