Apple introduced Apple Watch Series 12 on September 9, 2026, adding four Audio Intelligence features—Sound Recognition, Live Rewind, Siri Recap and faster Shazam—alongside a new Health Sensing System and S11 chip. The most substantial additions, Live Rewind and Siri Recap, are planned as beta features for late 2026 rather than being available as finished features at launch.

Apple says the audio functions do not create or store audio recordings. It also says raw audio is processed in a hardware-isolated Secure Exclave within the S11 chip and immediately deleted. Those are manufacturer claims, however, and the announcement does not provide independent testing of the privacy design, audio accuracy or summary quality.

Contents

What changed

The new watch uses its built-in microphone and processing system for functions that go beyond conventional notifications and voice commands.

Sound Recognition can detect environmental sounds such as sirens, alarms, doorbells and a baby crying, then alert the wearer. Apple says it can work when the iPhone is not nearby and describes the feature as intended to assist people who are deaf or hard of hearing.

Live Rewind displays the previous 15 seconds of a conversation as a text snippet after the user double-presses the Digital Crown. It is not described as a saved audio recording or a full transcript.

Siri Recap generates a title and high-level key points from a conversation for later review in the Siri app when enabled. A summary is different from a verbatim transcript: it compresses information and can omit or misinterpret details.

The fourth feature is faster Shazam, although Apple's announcement provides fewer technical details about how that improvement works.

Together, the features make ambient sound detection and limited conversation processing part of the advertised function of a mainstream smartwatch. The announcement does not establish that the watch continuously stores conversations or that every interaction is processed in the same way.

How the audio features work

A microphone converts sound into an electrical or digital signal that software can analyse for speech or environmental sound patterns. Apple describes Sound Recognition as using on-device intelligence, meaning the relevant computation is performed on the watch rather than sending the underlying sound directly to a remote service.

Apple presents Audio Intelligence more broadly as a combination of processing on the watch, Apple Intelligence models on an iPhone and Private Cloud Compute. The announcement does not specify how often Live Rewind or Siri Recap use the iPhone versus Private Cloud Compute, nor does it describe the processing path for every feature.

The distinction between the two conversation tools is important:

  • Live Rewind produces a short text snippet covering the preceding 15 seconds.
  • Siri Recap produces a title and high-level summary for later review.

Neither feature is described as identifying or attributing individual speakers. Apple says the system is designed not to do so.

Apple also says Live Rewind provides an audible chime, a full-display animation and a microphone indicator when activated. These signals are intended to make the action visible and audible to people nearby rather than making conversation processing imperceptible.

Users can opt in to each Audio Intelligence feature. Apple says Siri Recap can also be switched on or off through Control Center.

Privacy design and practical limits

Apple's privacy claims are central to the presentation of these features. According to the company:

  • Audio Intelligence features do not create or store audio recordings.
  • Raw audio is processed in a hardware-isolated Secure Exclave in the S11 chip.
  • The raw audio is immediately deleted after processing.
  • Live Rewind snippets and Siri Recap summaries are end-to-end encrypted in the Siri app and synced through iCloud.
  • The snippets and summaries are not accessible even by Apple.

A Secure Exclave is described by Apple as a dedicated compartment that processes audio separately from the rest of the system. On-device processing can reduce exposure of raw input, but it does not by itself establish that every associated output or piece of metadata remains local.

The release does not explain whether metadata associated with sound detection, feature activation, requests, diagnostics or synchronisation is retained or processed remotely. It also does not provide independent verification of the audio-deletion, encryption or inaccessibility claims.

The headline conversation features have significant deployment restrictions. Live Rewind and Siri Recap:

  • are scheduled to arrive in beta in late 2026;
  • require Apple Watch Series 12 or Apple Watch Ultra 4;
  • require an Apple Intelligence-enabled iPhone 16 or later;
  • exclude the iPhone 16e;
  • will initially be available in English;
  • will initially be unavailable in the European Union.

Apple also says features that rely on server-side models, including but not limited to Live Rewind and Siri Recap, are subject to daily usage limits. That means the practical experience will depend not only on the watch's hardware but also on a compatible iPhone, language and regional support, cloud-service availability and usage policies.

The announcement gives no accuracy figures or error rates for Live Rewind, Siri Recap or Sound Recognition. Until the beta features are available and independently assessed, it remains unclear how well they handle accents, overlapping speech, background noise, short utterances or complex conversations.

Health sensing and the evidence behind it

Apple Watch Series 12 combines the audio features with a new Health Sensing System and S11 chip. Apple says the watch measures heart rate every five seconds and measures heart-rate variability, or HRV, up to 24 times more often than before.

HRV describes variation in the time interval between heartbeats. It is influenced by multiple physiological and environmental factors and is not, by itself, a diagnosis or a direct measure of overall health. More frequent measurements may provide denser trend data, but the announcement does not establish that they produce better clinical decisions or health outcomes.

The watch also introduces a readiness score from 0 to 10. Apple says the score uses recent activity, training load, vitals and sleep score, and presents recommendations labelled Recover, Pace Yourself, Ready and Go For It. The company says the algorithm was developed using data from the Apple Heart and Movement Study with exercise scientists and physicians at Apple.

Apple reports that a study of more than 1,000 participants compared Apple Watch heart-rate accuracy with leading commercially available wearables and found that Apple Watch had the highest accuracy across the devices tested. The release does not provide the reference standard, numerical error estimates, confidence intervals, participant demographics, named comparator devices or a full protocol. No peer-reviewed publication or public study report is identified in the supplied announcement.

The evidence therefore supports a narrower conclusion: Apple is reporting a favourable result from a company-conducted comparison.

Apple says the Vitals app is for wellness purposes only and not for medical use. The announcement does not establish that the readiness score, HRV trends, Health Age or other generated insights diagnose disease or improve clinical outcomes.

Availability and what to watch

Apple Watch Series 12 is available in 42mm and 46mm sizes and starts at $399 in the United States. Apple says pre-orders began September 9, 2026, with store availability beginning September 18.

Apple lists India among the countries where the watch can be pre-ordered, with store availability also beginning September 18. The supplied announcement does not provide an India-specific price or confirm the availability of each Audio Intelligence feature in India.

The most important developments to assess after launch will be the independent performance of Live Rewind and Siri Recap, including transcription and summarisation errors; the final regional and language restrictions; the handling of metadata and derived outputs; and independent validation of the heart-rate comparison and readiness score.

Sources