Microsoft Patents AI That Rewrites Speech and Lips Live During Broadcasts

Microsoft Technology Licensing, the group's intellectual property arm, has filed a patent describing an artificial intelligence system to alter words and synchronize speakers' lips during live broadcasts. The document, revealed by the monitoring service Patentlyze and analyzed by Futurism, details AI agents capable of listening to speech in real time, identifying specific terms, and replacing or removing them before they reach the viewer's screen.

The mechanism operates in a near-real-time window: the AI detects the word, applies the replacement, and retouches the audio and video immediately before and after the altered segment to eliminate perceptible cuts. According to the analysis, replacement terms could even come from an internet search. Microsoft did not comment on the filing.

What the patent covers

The scope is broad. The document mentions conferences, presentations, award ceremonies, game shows, cooking programs, and sports rebroadcasts. Among the described uses are instant translation with lips synchronized to the viewer's language and the suppression of speech tics such as "uh," stutters, and regional expressions.

The list of eliminable words goes beyond noise: it includes mispronounced terms, slips of the tongue, offensive words, or words deemed inappropriate. Based on predefined parameters and each viewer's profile, the system could deliver personalized versions of the same live event — each person receiving a broadcast purged of what they don't want to hear.

The technical foundation already exists

The technology doesn't start from scratch. In April 2024, Microsoft Research Asia presented VASA-1, a model that animates a realistic face from a single photo and a voice file, with lips synchronized to the millisecond and natural head movement. According to the researchers' publication, the system runs in real time at up to 40 frames per second. At the time, Microsoft chose not to make it public.

The document also provides that the original audio and video be recorded and preserved as a "ground truth," allowing recovery of what was actually said. The patent, however, does not specify who would have access to this record or under what conditions.

Regulatory pressure in Europe

A patent is an intention: tech giants file thousands per year, and many never become products. Still, the timing is sensitive. In the European Union, the Artificial Intelligence Regulation has established, since August 2, 2026, transparency obligations for systems that generate or manipulate audiovisual content — including the need to flag synthetic or altered material.

The combination of real-time retouching with per-viewer personalization creates an additional layer of complexity: the broadcast ceases to be a single object and comes to exist in multiple simultaneous versions, each with distinct suppressions. Microsoft's document does not address how the system would handle the requirement to identify manipulated content.

The case also reignites the discussion about deepfakes in journalistic and institutional contexts. If the technology is brought to product, the central question ceases to be whether alteration is technically possible — it already is — and becomes who controls the button and how the public knows it has been pressed.