Microsoft has filed a patent for an artificial intelligence system that alters a speaker's voice and facial movements during live video feeds. The technology edits spoken audio to correct errors while matching lip movements on screen in real time. The concept remains an unconfirmed patent filing, with no timeline for an actual product launch.
The patent outlines software that monitors speech using automated AI agents. When a presenter stumbles, stammers, or uses filler words, the algorithm cuts the mistake and inserts corrected audio. The system simultaneously modifies video frames before and after the edit to ensure mouth shapes align with the revised sound. It can also strip out regional dialects, inappropriate slang, and offensive vocabulary during live delivery.
Microsoft envisions several practical scenarios for the tool, including instant language translation for international speeches. The engine can query web searches to generate replacement terminology on the spot. Because the software processes streams digitally, different viewers could receive separate customized edits of the same live event. Potential targets listed in the filing include academic lectures, sporting events, television programs, and corporate presentations.
Altering live statements presents clear verification risks. The patent addresses this by mandating that the system preserve the untouched original recording for legal comparison against the edited output. Microsoft has not confirmed whether it has built a functional prototype or if the concept exists solely on paper.
