Imperceptible live edits: What's behind Microsoft's app.
Microsoft has filed for a patent on an AI system for live broadcasts. It should almost immediately change the speaker's speech: find unwanted words, replace them with others, voice the replacement and adjust the lip movements to the new expression. The application was submitted in the spring of 2025 and published in the fall of 2026.
Delay - about 1 to 10 seconds. The system is designed for lectures, concerts, sports, streaming, podcasts, AR glasses, phones and cameras. Mentioned separately for simultaneous translation with lip-matching.
⁉ What should she be able to do?
Not only rude expressions can change, but also filler words, slang, offensive language, as well as expressions that the system considers inappropriate or offensive. New words can be found by searching the Internet. The app calls it quick fact-checking: the machine can correct the speaker if it decides they're wrong.
The most controversial thing is the customization for the viewer. The profile may take into account age, languages, region, culture, ethnicity, religion and education. Therefore, different people watching the same broadcast could theoretically hear different versions.
⁉ What's similar? Dubbing and Lip Sync Labs makes lip syncing close to real time. Hagen creates avatars with consistent facial expressions. D-ID claims very low latency. Vozo AI and Rask AI are involved in video translation and dubbing. Captions edit videos on the fly, Pinscreen edits lips and faces.
Voice and accent Sanas.ai softens the accent in real time, preserving the human voice. NTT changes the tone, pitch and rate of speech almost instantly.
Alibaba Cloud Broadcast Management Filters unwanted audio in many languages. RISM silences swearing and hides bad clips from streamers.
The point is not even that the machine is speaking in the name of man. Now she stands between the speaker and the audience and begins to decide what the gambler will reach the viewer.