Apple Watch beta tests 15-second speech recall and Siri conversation summaries

Apple’s watchOS 27.3 developer beta, released November 3, brings testing of two conversation-recall tools to Apple Watch: Live Rewind and Siri Recap. Apple’s support guidance, dated the same day, sets out what happens to the speech they process and the text they produce.

Live Rewind displays the preceding 15 seconds of speech after a double press of the Digital Crown, the watch’s physical dial. Siri Recap produces a title, summary and key points in the Siri app on Watch and iPhone. Apple’s examples include catching a missed restaurant remark or quickly delivered cooking instructions.

The tools belong to the previously announced Audio Intelligence suite. Apple’s product page still labels them as coming in beta late this year. 9to5Mac and The Shortcut report that both are testable in the new developer release. Apple’s cited pages give no firm date for general customer availability.

Each feature requires opting in. Siri Recap can follow a time-and-location schedule or run manually, with a Control Center switch. Live Rewind plays a chime from the watch speaker and shows an indicator even in silent mode or with headphones connected.

Unsaved Rewind text disappears about 30 seconds after the display dims; saving it or asking Siri continues its processing. Recaps can be set to auto-delete within seven days unless explicitly saved. Apple warns that generated text may be inaccurate or incomplete, and automated filters can miss sensitive information.

From the watch to the phone and cloud

Apple’s September privacy overview describes an encrypted transfer from protected hardware on the watch to its paired iPhone. The phone transcribes the speech and condenses it to less than half the original length. Apple says raw audio is deleted after processing, with no accessible recording file for the operating system or apps to share.

Private Cloud Compute, or PCC, generates the final Recap. Condensed text travels encrypted to the server with context that can include playing media, calendar information and broad location labels such as home or work. Apple says precise location and specific points of interest are excluded. Processing therefore extends beyond the watch.

Live Rewind initially performs speech recognition on the watch and iPhone. Saving its text to the Siri app or asking Siri about it sends the snippet to PCC. Audio retention, text persistence and server processing have separate rules.

Apple’s 2024 PCC security design requires personal data to be used only for the request and discarded afterward, including from logs and debugging material. It also excludes privileged operational access and makes devices send data only to servers running publicly listed software that researchers can inspect. These are Apple’s architectural assurances, not an independent audit of this beta.

Saved text and summaries collect in the Siri app. Apple says iCloud syncing is end-to-end encrypted with two-factor authentication and a device passcode. Users can edit, delete or export the text. Names mentioned in a conversation can appear in a recap, but the feature does not attribute statements to individual speakers.

Apple’s technical specifications require an Apple Intelligence–enabled iPhone 16 or later, excluding iPhone 16e, for these tools. Supported watches are Series 12 and Ultra 4. Server-model features have daily usage limits; the cited documentation gives no numerical allowance.

The November 3 US support page excludes users under 18. Initial language support is English, with the EU excluded from the initial rollout and China unavailable. Apple Intelligence and the Siri AI beta are required.