Apple has begun beta testing its Audio Intelligence capabilities for the Apple Watch, targeting developers through the iOS 27.3 update. The suite, first revealed at the company’s Surprise and Shine hardware event in September, adds four optional AI-driven tools for Series 12 and Ultra 4 owners. The features are being rolled out in a developer beta before a broader public test, and they aim to expand the watch’s contextual awareness while preserving user privacy.
The first two tools focus on ambient sound detection. A sound-recognition mode alerts deaf or hard-of-hearing users to nearby sirens, alarms or doorbells by displaying a visual cue on the watch face. A separate song-identification feature leverages Apple’s Shazam service to capture background music and present the track title and artist. Both functions operate locally on the device and do not retain raw audio recordings.
The more contentious capabilities involve live conversation capture. Live Rewind lets users double-press the digital crown to retrieve the last fifteen seconds of spoken dialogue as a text snippet displayed on the watch. An audible chime and on-screen animation signal that a recording has occurred. Users may dismiss the snippet, which then vanishes, or save it to the Siri app for later reference. An “Ask Siri” button beneath the text enables follow-up queries about unfamiliar terms or details.
Siri Recaps extends the concept by producing concise summaries of entire conversations. After a session, the watch generates a short sentence plus bullet points highlighting dates, deadlines, pricing or other factual items, without identifying speakers or storing raw audio. Users can enable the feature manually or schedule it for specific contexts such as work hours, and they may disable it at any time from the Control Center. Summaries are reviewed on the watch each morning, and any that are not saved self-destruct after seven days.
The underlying architecture relies on the S11 processor, an on-board microphone, and Apple’s secure enclave to hold a dedicated audio buffer. When the watch detects a conversation involving the wearer, the audio is transferred to the paired iPhone, where lightweight models strip filler words and encrypt the condensed text. This data is sent to Apple’s Private Cloud Compute for brief summarization, then immediately deleted. The resulting recap returns to the iPhone, remains encrypted during sync, and is inaccessible to Apple.
Apple emphasizes that the new tools do not affect battery life on either the watch or the iPhone, and that users retain full control via the Control Center or scheduled settings. Recaps and Live Rewind snippets are stored only on the user’s devices, encrypted in transit, and automatically removed after a week if not saved. The company plans to expand the beta to a public test before releasing the features to all Apple Watch owners.