Microsoft 專利申請描述可即時修改直播語音並同步重繪唇形的 AI 系統
Microsoft Patent Application Describes AI That Edits Live Speech
Microsoft 提交了一項 AI 系統專利申請,描述可在直播影片中替換或刪除説話者的用詞,並重繪唇形以配合修改。申請編號為 US 2026/0341309 A1,於 2025 年 4 月提交、2026 年 10 月 29 日公開;系統按規則及觀眾年齡、語言或所在地等個人資料選擇要處理的字詞,並以約一至十秒延遲播放結果。
Microsoft has filed a patent application for an AI system that can change what someone says in a live video and redraw their lips to match. The filing, US 2026/0341309 A1, was submitted in April 2025, published on October 29, 2026, and first spotted by patent tracker Patentlyze. In short, an AI agent scans the live audio and picks words to swap or remove based on rules and a viewer's profile, such as their age, languages, or location. A generative model then replaces those words in both the audio and video, reworks the frames around each edit and plays the result with a delay of roughly one to ten seconds. It could run locally on AR glasses, phones, tablets or cameras.
A separate claim covers live translation with matching lip movements, which is easily the most practical use. However, the filing also lists filler words, stutters, mispronounced or offensive words, and colloquial terms like "y'all" or "bite the bullet" as things to replace, and the new words can even come from an internet search. It mentions lectures, award shows, game shows, and live sports as examples. The original feed may be stored as "ground truth," but the main claim says nothing about telling viewers what was changed. Of course, Microsoft files plenty of patents that never become products, and the examiner has already rejected the claims as filed, which is routine at this stage. Sony has been busy with AI patents as well, ranging from frame generation to an AI that can play for distracted players.
來源:TechPowerUp · techpowerup.com


