Microsoft has filed a patent application for an artificial intelligence system that can change what a speaker says during a live broadcast and adjust the speaker’s mouth to match the new words. The Microsoft AI patent describes software that listens to live audio, picks out words to remove or replace, and rebuilds both the sound and the picture within seconds. Computerworld reported the filing on 4 November 2026, after the patent tracker Patentlyze spotted the published application.
The Microsoft AI patent is US publication 2026/0341309 A1, titled “Near Real-Time Artificial Intelligence (AI) Enhancement of Video or Augmented Reality (AR) Presentations of Live Events.” Microsoft Technology Licensing filed it on 29 April 2025, and the US Patent and Trademark Office published it on 29 October 2026. Publication happens automatically, about 18 months after filing, and it does not mean the office has granted the patent.
What the Microsoft AI patent describes
An AI agent scans the audio as it arrives and flags words to change, using a set of parameters and a profile of the viewer. A generative AI system then replaces the chosen words in the audio and the video, reshaping the speaker’s mouth and smoothing the frames around the edit so the lips match the new speech. The Microsoft AI patent says the system could run on phones, tablets, cameras and AR glasses with on-device chips.
The filing defines near real time as roughly one to ten seconds, close to the delay broadcasters already use so a person can filter profanity by hand. That short window is what makes the Microsoft AI patent different from ordinary editing. The change would land while the event is still on screen.
Uses named in the Microsoft AI patent
Microsoft lists translation first. Speech in one language could be replaced with words in the viewer’s language, with the video changed to match. The filing also names filler words such as “um,” “uh,” “like” and “you know,” along with stutters, mispronunciations and slang. A viewer could blacklist words or phrases, and the system would strip them out.
The settings named in the filing are lectures, award ceremonies, sporting events and live television programmes. Microsoft also says the system could remove profanity or offensive language to suit a viewer’s age, language or location. The original audio and video would be stored as a record of what was actually said.
Personalisation is built into the design. Two people watching the same event could hear different words, because the edits follow each viewer’s profile. That raises a separate question about shared events. A live broadcast has long been one feed that every viewer sees at the same time, and the filing describes a feed that can differ from one screen to the next.
Why the Microsoft AI patent raises questions
The Microsoft AI patent raises an obvious problem. A tool that can put new words in a speaker’s mouth could let a viewer watch a public figure appear to say something the person never said, in footage that looks untouched. The filing appears to accept that risk. One claim proposes keeping the original recording as ground truth, so the unedited version still exists.
The patent tracker that flagged the filing says a US examiner has already rejected the claims as written. That is a routine first step, and Microsoft can answer by amending the claims. Futurism reported that Microsoft did not immediately respond to a request for comment. Computerworld noted that it is unclear when, or if, Microsoft will launch the technology.
Microsoft has worked on related ideas before. Its VASA-1 research project, published in April 2024, showed how AI could turn a still image and an audio clip into a talking-face video. The patent application goes further, because it covers live speech rather than a recording. For now, the system exists only in the filing.
The application will now move through the patent process, where Microsoft can amend the claims and argue for them. Nothing in the Microsoft AI patent sets a date for a product, and Microsoft has not said it plans to build the system. Until it does, the idea remains a description of what the company wants to protect, not a tool anyone can use.