Pink noise cost sleepers 19 minutes of REM in a Penn lab

Somewhere near a lot of beds tonight, a small speaker will be switched on to hiss. It might be a dedicated machine, a phone app or a ten-hour video. According to the University of Pennsylvania, white noise and ambient podcasts now account for 3 million hours of listening a day on Spotify alone.

The pitch is simple. A steady, featureless sound covers the car door, the neighbor and the 5 a.m. flight, so the sleeper never notices them. It is one of the most widely used sound interventions there is, and the evidence for what it does to the sleep underneath has always been thin.

A Penn team did, and the result, published this year in the journal SLEEP, is not what the packaging says.

What was tested

Mathias Basner, a professor of sleep and chronobiology at Penn’s Perelman School of Medicine, and 15 colleagues brought 25 healthy adults, ages 21 to 41, into a sleep laboratory for seven consecutive nights. None had a sleep disorder, and none was in the habit of sleeping with sound on.

Each night offered eight hours in bed with full polysomnography, the electrode recording that separates light sleep, deep sleep and REM. The nights differed only in what the room sounded like:

• A silent control night.
• A night of intruding noise: 93 events, among them jets, helicopters, drones, a car, a train and alarms, peaking between 45 and 65 dBA.
• A night of nothing but steady pink noise at 50 dBA.
• The intruding noise with pink noise underneath it, at 40 dBA and again at 50 dBA.
• The intruding noise with foam earplugs, which cut the sound reaching each participant’s ears by an average of 25.5 dB.

Pink noise is the deeper cousin of white noise. Both contain every audible pitch at once, but pink noise carries more of its energy in the low frequencies, which is why it reads as steady rain where white noise reads as radio static. ## What happened

The intruding noise did what you would expect. Against the silent night, it took 23.4 minutes out of N3, the deepest stage of sleep.

Pink noise on its own, with nothing to mask, took 18.6 minutes out of REM, the stage in which most vivid dreaming happens.

The combination is the finding that matters, because it is how people actually use these machines. Adding pink noise under the jets and trains did slightly reduce how often those events broke up sleep. But the nights as a whole were worse than the intruding noise alone: deep sleep and REM together fell by 42.6 minutes with the quieter pink noise and by 44.8 minutes with the louder, and Penn reports that participants spent 15 more minutes awake.

Earplugs went the other way. They returned 16.9 of the 23.4 lost minutes of deep sleep, about 72% of it, and in the authors’ summary “mitigated nearly all” of the noise’s effects on sleep. They began to fail only against the loudest events, at 65 dBA.

Why REM is the worry

The authors’ concern is about what the lost stage is for. Their paper cites “the importance of REM sleep for memory, emotion regulation, and neurodevelopment” and concludes that the result should “caution against the widespread and indiscriminate use of broadband noise.”

Basner put the sharpest point on children. “REM sleep is important for memory consolidation, emotional regulation and brain development,” he said in Penn’s release, “so our findings suggest that playing pink noise and other types of broadband noise during sleep could be harmful—especially for children whose brains are still developing and who spend much more time in REM sleep than adults.”

That sentence deserves a careful reading. No child took part in this study. The warning about nurseries is an inference from two facts, that pink noise reduced REM in adults and that infants spend far more of their sleep in REM. It is a reasonable thing to worry about and it has not been measured.

Was the evidence ever there?

This is not the first time the same group has looked. In 2021, Basner and three colleagues reviewed 38 studies of continuous noise as a sleep aid in Sleep Medicine Reviews. The results ran in every direction, from better sleep to worse, and the reviewers rated the quality of evidence that noise improves sleep as “very low,” which, they wrote, “contradicts its widespread use.”

The best case for the machines comes from hospitals. A Mayo Clinic team presented a review at a 2025 hospital-medicine conference covering seven randomized trials and 496 hospitalized adults, six of them using white noise. It found evidence of improved sleep efficiency, and only 8 participants quit because the sound bothered them.

Those two findings are easier to reconcile than they look. A hospital ward is a loud place, and several of those trials measured sleep by questionnaire. A steady sound may well make a chaotic night feel better. The Penn study asked a different question, with electrodes, about what happens to the structure of sleep, and found a cost that the sleeper cannot perceive directly.

The honest read

This is a small, short study, and three limits matter.

It enrolled 25 young, healthy adults who do not normally sleep with sound. People who have used a machine for years may have adapted, and the study cannot tell us. Each condition lasted a night, not a season. And it tested one color of noise at two levels, so it says nothing certain about white or brown noise, or about a fan across the room.

It also has an unusual strength for this corner of sound research: objective sleep staging, a silent control, and a head-to-head against a cheap alternative. The funder, the US Federal Aviation Administration, has no sound machines to sell.

We owe a correction of emphasis ourselves. Our report on one night of road noise listed a white-noise machine among the cheap fixes for a noisy bedroom. A machine adds sound to the room; it does not take any away. On this evidence the earplugs on that list are the better-supported choice, with the plain caveat that whatever blocks a jet also dulls a smoke alarm or a child calling out. Penn puts the share of Americans who already sleep in earplugs at about 16%.

If a sound machine is the only thing that gets you to sleep, one lab week in 25 strangers is not a reason to panic. It is a reason to notice that the claim on the box was never tested the way this was.

What to watch

The paper names the missing studies itself: long-term use, other colors and levels of noise, and “vulnerable populations,” meaning above all infants and toddlers, whose parents are the machines’ most devoted buyers. A sleep-lab study of habitual users, comparing nights with their machine on and off, would answer the adaptation question quickly. We will report it when one appears.