What the system could change
The patent lists several uses, from practical translation to edits that raise harder questions about who gets to shape what an audience hears.
The system may use internet searches to find replacement words. The original video and audio could also be recorded and stored as a “reference recording,” according to the patent.
Examples of possible settings include lectures, presentations, award ceremonies, game shows, cooking shows and sports broadcasts.
The edit is the experience
Video and audio editing are old tools. The change here is that AI agents could apply them automatically during a live event, making the edited version feel like the event itself. Instead of turning off a broadcast, a viewer could have someone else’s speech adjusted to suit their taste.
I think the more consequential question is not whether the system can remove filler words. It is what happens when its idea of an error reaches into dialect or phrasing. A tool that cleans up speech could also make a speaker sound less like themselves, while giving each viewer a different version of the same moment.
Microsoft did not immediately respond to a request for comment. The patent says a reference recording could help address misuse and confusion, but it leaves open how viewers would know what had been changed. A live event that can be quietly tailored to each audience may still be live in timing; whether it remains a shared experience is another matter.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X