
Microsoft filing describes per-viewer rewrite of live speech and lip movement; no product is announced
The primary event is a Microsoft Technology Licensing patent application, not a deployed speech-rewriting product. Reported claims would let a viewer profile drive word substitution and lip regeneration, with the unaltered feed preserved only in a dependent claim. Employment, privacy, and communication effects follow only if that option is exercised; the file sets no such date.
Microsoft Technology Licensing filed US application 2026/0341309 A1 on 29 April 2025, listing David Callaghan and Dinesh Kumar as inventors. The application, published 29 October 2026, describes near-real-time alteration of live video or augmented-reality presentations. Reported claim language has an agent select spoken words to modify from parameters tied to a viewer profile, replace the audio, and generate frames so the mouth matches the new words. Status on the record is an application under examination, not a granted patent and not a shipped service. Microsoft has announced no build plan and did not answer a Futurism comment request.
Reports of the file describe a viewer profile that may hold age, language, region, religion, ethnicity, education, and voice or video samples, so two viewers could hear different wording of one feed. Stated functions cover filler and stutter removal, slang replacement, translation, and profanity filtering, inside a one-to-ten-second delay. Replacement terms may come from an internet search. The text, as reported, does not call that fact-checking; the label is an outlet inference. Only one dependent claim is described as keeping unaltered audio and video as ground truth. An office action dated 29 July 2026 rejected the claims as filed.
Secondary coverage treats the filing as a near-term intervention in viewers’ jobs, privacy, and online speech. The application states neither a ship date nor a product. If the reported claim tree is accurate, personalized word substitution sits in the main claim and retention of the original in one dependent claim. Nothing reported requires notice that a feed was altered. The counterparty is the speaker, whose words would cease to be identical across audiences, and any institution that treats a broadcast as one record. Profile fields would be a data input only if the system were built.
Microsoft’s gain is option value: a priority date beside Teams Interpreter, which translates audio and does not rewrite the face. Tbreak reports a Sony filing for rule-based replacement of sensitive game moments; a parallel application, not a joint product. Costs on the ledger are method disclosure, the July 2026 rejection, and labeling exposure only if a product followed. The named functions correspond to delay-desk editing, interpretation, and compliance cuts. Those roles change only if Microsoft exercises the option. A twelve-month shift in jobs or everyday speech is outside the document. Next signal is an amendment or abandonment, not a launch announcement.
Sources (2)
- [1]US patent application 2026/0341309 A1(https://patents.google.com/patent/US20260341309A1)
- [2]Microsoft Wants AI To Put Words In Your Mouth - In Realtime(https://www.zerohedge.com/political/microsoft-wants-ai-put-words-your-mouth-realtime)