The sound machine hums all night. Your REM sleep, quietly, does not.

It's 10:47pm. Your teeth are brushed. Your phone is charging in the other room, exactly where you promised yourself you'd leave it. The last thing your hand does before the day ends is the tap you've made a thousand times — on the little white cylinder by the bed, or on the app that plays pink noise through the speaker across the room. The room fills with a low, soft wash of something between rain and a refrigerator hum. You close your eyes. This is the ritual. This is how you sleep.

You are not doing anything wrong. You bought the machine because sleep felt hard, or because the neighbour above you keeps hours you don't, or because the highway is a block away, or because it worked the one week you tried it in your twenties and you never took it back off. It's the one small helper the internet has agreed is safe.

The internet has been talking about sound machines for years. The sleep lab has only recently caught up.

The ritual has a cost the EEG can see

In February 2026, a team led by Mathias Basner at the University of Pennsylvania published the first careful in-lab test of what a night of pink noise actually does to sleep architecture. They put 25 healthy adults, ages 21 to 41, through seven consecutive nights of polysomnography — full EEG, eye movement, muscle tone — in acoustically isolated bedrooms. Each night was a different acoustic condition: a silent control, simulated traffic and aircraft noise (93 events at 45–65 dB), pink noise alone at 40 or 50 decibels, pink noise layered over the traffic noise, and traffic noise with foam earplugs. ^[1]

The finding that made the sleep community pay attention was not about traffic. It was about the fix.

Continuous pink noise at 50 decibels — roughly the volume most sleep apps and bedside machines default to — was associated with about a 19-minute reduction in REM sleep per night compared to the silent control (p < .001). ^[1] In a night that typically contains only 90 to 120 minutes of REM, losing 19 of them is not a rounding error. It is close to a fifth of the sleep stage responsible for emotional processing, memory consolidation, and dreaming.

The traffic noise, unsurprisingly, cost the participants something too — about 23 fewer minutes of N3, the deep, slow-wave stage that does most of the physical restoration in a night. What was surprising was what happened when pink noise was layered over the traffic to mask it: both stages were still shortened, and participants spent an additional 15 minutes awake. ^[1] The intervention meant to rescue the night made it a slightly worse night.

Earplugs did what the sound machine could not

The clean part of the paper is the comparison arm. When the same traffic noise was paired with foam earplugs instead of pink noise, the participants' N3 and REM sleep came back to roughly the levels of the silent-control night. ^[1] The mask that removed sound protected the architecture. The mask that added sound did not.

An earlier crossover pilot from the same group, published in Communications Medicine in 2026, had already flagged that intermittent traffic noise fragments sleep and shifts the blood metabolome the next morning — even when total sleep time looks preserved. ^[2] The follow-up in Sleep was what named the mechanism: it is not the amount of sound in the room. It is that the auditory cortex does not turn off when you do. Continuous input, even the soothing kind, keeps something engaged that the REM cycle needs quiet to build.

Why it matters for the felt experience

If you have ever had a full eight hours next to a running fan or a lulling app and woken up flat — a little short with the first message you answered, a little slower to laugh at something your partner said before coffee — you may have felt the quiet edge of this without knowing what it was.

REM does the work you notice most on the days you don't get it: mood regulation, emotional colour, the shape of your patience. A commentary published alongside the Basner paper made the population version of the point. In Europe alone, environmental noise pollution is estimated to cost more than 900,000 disability-adjusted life years each year, largely through sleep fragmentation and its downstream effects on the heart. ^[3] Most of us are not in that severe tail. But the smaller version is real too, and it hides. You don't notice the missing REM the way you notice the missing hour. You notice that the days feel a half-step flatter than they should.

For anyone who works from where they sleep, the exposure lengthens. There's no office to break up the acoustic dose. The white-noise app that goes on at 10:47pm is often still going at 7am when the first meeting begins. The room's soundscape is now a background you live inside.

What to do tonight

You don't have to give up the sound. You can give it a boundary.

Three small moves, in order of how much sleep science they buy back:

• Put the machine on a timer. Most sound machines and every white-noise app can run for 30, 60, or 90 minutes and then stop. If you use the noise only to fall asleep, that's all it needs to do. Once you're out, the room can be silent, and your REM cycles can build against a quiet floor.
• Turn the volume down. Basner's team saw an effect at 50 decibels. Their lower dose was 40. If the machine has to stay on all night for a real reason — a snoring partner, a night-shift neighbour — set it as low as you can still tolerate and cover the last ten to twenty minutes of the night with silence if you can.
• Try earplugs for one week. If the reason you started the machine was to mask the world outside, foam earplugs are the intervention the trial actually endorses. In the paper's data, they were the only condition that kept both deep sleep and REM close to the silent night. Cheap, disposable, and quiet in a way a machine can't be.

If the quiet feels loud at first, one round of slow-exhale breathing is a good bridge. Inhale for four. Hold for seven. Exhale slowly for eight. The long exhale is what shifts the nervous system from scan the room into release the day.

The reframe

The sound machine is not a villain. It is a scaffold. Some scaffolds hold buildings up. Some sit there long after the wall is finished, quietly costing rent.

For years the cultural script around sleep sounds was a yes-or-no question — does it help me fall asleep, or doesn't it. The 2026 evidence is quietly retiring that question and asking a better one: what kind of night does it leave me with. Usually a night that counts as sleep. Sometimes a night with a slightly shorter version of the stage that makes the next morning feel like yours.

The lever is small and it is entirely yours. A timer. A lower dial. Or a pair of earplugs and the room's own quiet.

Tonight

Find the sleep-timer setting on your machine or app. Set it to 60 minutes. If you don't know where the setting is, look now — three taps, and the ritual becomes a first hour instead of a whole night.

Tomorrow morning will tell you whether it was worth it. We think it will be.