Apple Health hearing data: exposure samples and noise events

Two exposure streams, two questions

HealthKit defines environmentalAudioExposure for sound exposure in the surrounding environment and headphoneAudioExposure for exposure from headphones. The first describes the acoustic environment around the person; the second describes listening through connected headphones. A daily total that combines them without labeling the source hides the difference between the two experiences.

Apple's iPhone guide also notes that headphone measurements are most accurate with Apple or Beats headphones; audio from other headphones may be estimated from the iPhone volume. Preserve the device and source-app metadata so a model does not present every value as the same kind of direct measurement.

• Environmental exposure: sound around the wearer, with Apple Watch collection available.
• Headphone exposure: audio delivered through connected headphones.
• Source and measurement context: required before comparing or aggregating samples.

A quantity sample is not a notification event

The environmentalAudioExposure quantity uses sound-pressure units and represents an equivalent continuous level over an interval. The environmentalAudioExposureEvent category records that Apple Watch generated a noise notification. One is a quantitative exposure sample; the other is a discrete system event.

HealthKit documents the event as read-only: an app can request permission to read it but cannot save a new environmental exposure event. Do not count events as if they were minutes of exposure, and do not treat the absence of an event as proof that the environment was quiet or harmless.

What a Noise notification means

Apple says Apple Watch periodically measures environmental sound while it is being worn. If the average sound level over three minutes reaches or exceeds the chosen threshold, the watch can notify the person. Apple also states that the watch does not record or save the sounds for this feature.

That event is therefore evidence about a threshold rule and a time window, not an audio file or a complete exposure history. Threshold settings, wear time, watch model, and gaps in collection should remain attached to an analysis of notification events.

• The rule uses a three-minute average, not a single instantaneous peak.
• The threshold is configurable and may vary by feature or region.
• No saved audio recording is created by the Noise notification measurement.

Coverage and source change the interpretation

A blank period may mean the watch was not worn, Noise notifications were off, the iPhone was not connected, a compatible headphone was unavailable, or a source app did not write to HealthKit. It is an unknown interval, not a quiet interval. A delayed sample from a secondary iPhone paired with Apple Watch also needs to retain its acquisition context.

For a trustworthy trend, group records by source and device before calculating duration or exposure summaries. Keep overlapping samples visible until the deduplication rule is explicit, and do not merge environmental and headphone records merely because their timestamps overlap.

A synthetic hearing-data timeline

The following values are invented and contain no real audio or health data.

• Monday 09:00–09:20 — environmental exposure samples from Apple Watch; no notification event.
• Monday 18:00–19:00 — headphone exposure from a supported headset; source metadata retained.
• Tuesday — a noise notification event was recorded; the event does not include saved audio.
• Wednesday — no samples because the watch was not worn; the environmental exposure is unknown.
• Defensible summary: the records show different sources and one notification event; they do not establish hearing ability or damage.

What an AI analysis should preserve

Before asking a model to summarize hearing data, include the HealthKit type, timestamp and interval, sound-pressure unit, source app, device and headphone context, notification threshold when known, authorization scope, selected date range, missing intervals, duplicate handling, and extraction time.

The model can describe exposure by source, compare two declared windows, list notification events, and flag incomplete coverage. It should not infer hearing loss, convert a notification into a diagnosis, invent a population normal range, or imply that no event means no risk.

MCP may be an optional third-party interface for a minimized, user-approved summary. It is not an Apple protocol, does not grant HealthKit permission, and should not silently transmit raw location, free-text notes, or lifetime listening history when an aggregate is sufficient.

Privacy and medical boundaries

Listening habits, locations, device identifiers, and noise-event timing can reveal work, travel, and social routines. Request only the fields needed for the question. Prefer source-labeled aggregates over precise routes or minute-by-minute records when those details are unnecessary, and disclose external model processing, retention, deletion, and revocation behavior.

This article is a data-model and interpretation guide. It does not diagnose hearing loss, assess clinical hearing thresholds, recommend treatment, or replace an audiologist or clinician. Persistent hearing concerns should be evaluated through an appropriate professional service.