A new Microsoft patent application describes an artificial intelligence system that could change a person’s words as they speak in a live video. It would also adjust their lip movements so the change goes unnoticed. The system could automatically translate speech into other languages, correct mistakes and remove unwanted expressions within secondsof them being spoken.
The application was filed on 29 April 2025 and published on 29 October 2026, according to Gadget Review. For now it is a technology design, not an available product, and it is not known whether or when Microsoft intends to release it.
How the system would work
According to Microsoft’s description, an AI system would monitor the audio and video of a live broadcast and identify words or phrases that can be modified.
It would then replace those words, adjust the audio and generate new video frames that change the movements of the speaker’s mouth and lips. Image and sound would be modified together, so the viewer sees the speaker naturally pronouncing the new words.
Translation and speech editing
For example, a person could watch a lecture in English and hear it in Greek, while the speaker on screen appears to be speaking Greek.
The system could also remove hesitations and repeated words, correct mispronunciations and replace local expressions with more common ones.
The application also allows profanity or offensive remarks to be removed, depending on the viewer’s preferences. Two people watching the same live broadcast could therefore hear different versions of what the same speaker says.
Correcting inaccurate information
One of the most ambitious capabilities in the application is the correction of inaccurate information. The system would search for information online and replace an incorrect statement with the wording it considers correct.
This does not mean, however, that AI would always be able to tell truth from error. An incorrect “correction” could distort the meaning of a statement without the viewer realising that the words had been changed.
Questions over the reliability of live broadcasts
The ability to alter image and sound almost simultaneously raises serious questions about the reliability of live broadcasts.
Such technology could be used to attribute statements that were never made to politicians, journalists or business executives, creating highly convincing manipulated videos known as deepfakes.
The issue is not limited to malicious use. Even when changes are meant to improve communication, the question remains of who decides which words are removed or replaced.
If, for example, an application automatically removes terms it considers offensive or corrects a politician’s statements, viewers may not know what was actually said.
Safeguards described in the application
Microsoft’s application includes the option of storing the original, unedited video, so that a reference copy of what was said exists.
Keeping the original material does not in itself ensure that viewers will be told about the changes they are seeing, or that the speaker will have consented to them.
Potential uses
The technology could be used in international meetings, video conferences, educational courses and other live broadcasts, making communication easier for people who speak different languages.
It could also make speeches with frequent errors or pronunciation difficulties easier to understand, without the video having to be edited after recording.
At the same time, it highlights the need for clear public disclosure when AI is used to modify speech and video.
With information from Gadget Review
