My Apple Watch showed elevated breathing disturbances. What could I check next?

A reading worth looking into

Last night I wore my Apple Watch and ran SomniSense on phones beside the bed. The two phone reports shown here gave a Breathing Irregularity Index (BRI) of 22.1 on iPhone 14 and 20.7 on Pixel 7.

Seeing the readings in a similar numerical range gave me more confidence in the work I am doing. This is one real night from my own use. Apple’s breathing disturbance value and SomniSense’s BRI have different definitions, recording periods and calculation bases.

If your watch has ever shown an elevated reading, you may have had questions like these: When did it happen? Was it spread across the night or clustered in a few periods? What did the breathing sound like?

The two Apple Health screens I saved show the level, value, recording period and a view across nights. They do not show individual moments within this night or a sound recording to open. Those were the details I wanted to explore next.

From a number to a moment

In SomniSense, the Pixel report put 150 breathing irregularity marks back into the night from 23:12 to 07:06. The iPhone report showed 162 marks. Under the summary, there was somewhere to keep looking.

For example, the Pixel’s minute view at 00:12:48 showed a waveform with a marked region. Its detail text described a model-marked shallow-breathing candidate from 00:12:42 to 00:13:11—an estimated 29 seconds—and explained that the interval crossed a minute boundary.

The playback control is there so a user can listen to the sound before and after a marked interval when the audio is available.

This is the kind of detail I wanted the app to make accessible. A concern about a whole night becomes a time and a record you can revisit. Another night lets you ask whether similar marks recur, without having to remember hours of sound or rely only on how you feel in the morning.

How can a phone do this?

The breathing-audio research behind SomniSense used audio samples paired with polysomnography (PSG) labels for training. Our research preprint describes 80 person-nights from 40 participants, including 2,953 paired 200-second audio windows.

Our work also adapts one-dimensional Coordinate Attention to combine broad and local audio features while preserving information about where patterns occur in time. That research foundation matters to me: the goal is to turn a night of sound into patterns worth reviewing, then make the surrounding evidence easy to find in the product.

You can inspect the methods and evaluations in our research and evidence pages. The study evaluates audio-classification tasks; this personal night adds a different kind of observation. Neither makes the phone’s marks a clinical diagnosis.

A concrete place to start tonight

If you already wear a watch, SomniSense can add a sound record you can inspect and, when available, replay. If you do not use a wearable, you can start with a compatible phone you already own.

That is what I wanted to share about this night: a way to follow the question, “What happened while I was asleep?” Put your phone beside the bed tonight. In the morning, see whether there is a moment you want to understand better. Keep recording if you want to see what repeats.

This is the founder’s own use of SomniSense, not an independent customer review. The phone candidates have not been aligned event by event with Apple Watch. SomniSense analyzes sound and marks possible breathing irregularities; clinical questions belong with a qualified health professional.