Hush by Sky Wolf Studio, a free sound machine and sleep sound player

drag the ring to set volume

18%

Tonight

Night music

Stop after

Programs

Sounds

Made, not recorded

Level42%

How often45%

Tempo70 bpm

These are generated live, note by note. There are no audio files in this app. That is why it is one page, works offline, and never repeats itself.

The tempo range is not arbitrary. Trials of music for insomnia keep landing on the same recipe: slow tempo between sixty and eighty beats a minute, one instrument, soft and smooth, simple structure, thirty to forty-five minutes before bed. The tempo slider only moves inside that window on purpose. Jespersen et al., Cochrane review, 2022 update (13 studies, 1007 people): listening to music improved self-rated sleep quality by about 2.8 points on the 21 point PSQI scale (10 studies, 708 people), moderate certainty evidence. Later reviews converge on the 60 to 80 bpm figure.

Toneflat

Widthwide

Left is darker and deeper, right is brighter and airier. Same noise, different weather. Width spreads the sound between the ears: each side gets its own noise, which reads as air in the room. Turn it to mono for a single speaker across the nursery.

Target frequency440 Hz · A4

Amount0 dB

Widthwide

Golden stack

Adds resonances at 1.618× and 2.618× the target. Irrational spacing, so they never beat against each other.

Boost finds a pitch inside the noise. Notch removes one, and that is the setting used for tinnitus masking, tuned to the tone you hear.

Mix0%

Rate0.15 Hz

Depth45%

Resonance30%

Lock to target

Freezes the sweep and tunes the delay so the noise rings at your target pitch. Turn resonance up to hear it.

Golden drift

Three modulators at 1×, 1.618× and 2.618× the rate. φ is irrational, so the pattern never repeats and the texture never settles into a loop your ear can learn.

Speed72 bpm

Amount50%

Heartbeat is a filtered low thump at a resting maternal rate. Shush is the rhythmic hiss caregivers make instinctively. Waves is a slow swell. Breathe swells at 5.5 cycles a minute, so you can pace your own breathing to it.

Texture level50%

Texture is what makes a sound itself instead of another shade of noise: the womb's blood-flow swoosh pumping at the heartbeat's own rate, an engine under the road with cars that occasionally pass, raindrops that actually patter, waves that build and break, the flutter of fan blades. Generated live like everything here. No recordings.

Slow-wave bursts

Short pink-noise bursts at 0.8 per second, in alternating on and off blocks. Built to the stimulus described in the published protocol rather than an approximation of it.

Burst rate0.80 Hz

Block length10 s on / 10 s off

Burst level35%

What this is copying. In lab studies, short pink-noise bursts played during deep sleep increase slow-wave activity, and in some studies improve next-morning word recall. The strongest results come from closed-loop systems that read EEG and fire on the up-phase of each brain wave. Open-loop versions, meaning no EEG and fixed timing, which is all a phone can do, have also raised the share of the night spent in N3 in a small supervised study. Schade et al., Nature and Science of Sleep 2020 (n=8, supervised, in-lab): 50 ms pink bursts, 5 ms ramps, one every 1.25 s. Papalambros et al., Front Hum Neurosci 2017 (n=13) for the closed-loop version.

Small studies, lab conditions, EEG supervision. A phone in a bedroom is not that. Treat this as a faithful copy of a research stimulus, not as a device that will reproduce a research result. Set the burst level low, because the studies used levels near the listener's hearing threshold, not loud ones.

Everything here runs on your device. The microphone is only read to measure loudness. No audio is recorded, stored or sent anywhere. Turn it off any time and the app works exactly as before.

Use the microphone

Turns on the loudness meter and adaptive masking.

— dB, estimated

peak

3050 · nursery reference85

Turn the microphone on, put the phone where the baby's head is, and read the number.

Calibration+100 dB

Phone microphones are not calibrated instruments and every model differs, so this is an estimate. If you have a sound level meter, or a calibrated app on another phone, the button above will play a steady sound and work the difference out for you. Without one, treat this as a relative reading: good for comparing one spot in the room against another, not for a measurement anyone should rely on.

Adaptive masking

Keeps the noise low and only lifts it when the room actually gets loud: a truck, a door, a sibling. Steady sound stays out of the room the rest of the night.

Sensitivity6 dB over baseline

Lift+8 steps

Adaptive masking compares a fast average against a slow one, so your own steady noise settles into the baseline and doesn't chase itself upward.

Ok-to-wake light

In full screen, the circle stays red until the time you set, then turns green. Toddlers understand it long before they can read a clock.

Sunrise

The screen warms and brightens over the half hour before the set time, and the sound thins out with it.

Both need the phone left in full screen with the screen kept on. Plug it in.

—:—

Fade out over60 s

Keep the screen on

Stops the phone sleeping while sound is playing.

A timer is the default here on purpose. See the evidence panel below for why.

Headphones only, and not on babies or young children. Binaural beats need one ear per channel to work at all, and headphones make it far too easy to reach unsafe levels on small ears. On speakers this is just a quiet drone.

Carrier160 Hz

Beat4.0 Hz

Level0%

Isochronic instead

One tone pulsed on and off rather than two detuned tones. Works on speakers.

Evidence for brainwave entrainment is thin and mixed. Treat this as a texture you might find pleasant, not a treatment.

This runs a real experiment on you. Pick two sounds. Each night the app secretly picks one of them, balanced so you get an even number of each. In the morning you log how the night went, by hand or by importing from your tracker. When there are enough nights, you get an answer. Until then it says so.

Sound A

Sound B

Judge it on

Keep it blind

Hides which sound is playing until the trial ends. Strongly recommended, because knowing which one is "the good one" is enough to change how you sleep and how you rate it.

Nights per sound14 each · 28 total

How did it go—

Time asleep Deep sleep REM Minutes to fall asleep Times woken Sleep score

Fill in whatever your tracker gives you. Anything you leave blank is simply ignored. If you only ever fill in the first slider, that still works.

Files, not accounts. There is no honest way for a website with no server to log into your watch account and keep your password safe, so HUSH does not try. Instead you export from your tracker and drop the file here. It is read on your phone, never uploaded, and the file itself is not kept.

Import a file

Where to find the export.
Apple Health: Health app, tap your picture, Export All Health Data. Unzip it and pick export.xml. It is a big file; this reads it in pieces so it won't run your phone out of memory.
Oura: the web dashboard has a CSV export under your account settings.
Fitbit: Account settings, Data Export. Use the sleep JSON files.
Samsung Health, Garmin, Withings, Whoop: all export CSV. Drop it in and HUSH will try to work out the columns.
Anything else: any CSV with a date column and a sleep column will import.

Live heart rate

Not connected

Works on Android and desktop Chrome only. Apple does not allow web pages to talk to Bluetooth on iPhone or iPad, and the Apple Watch does not share heart rate with anything outside Apple's own apps. That is Apple's decision, not something an app can work around. Chest straps and most Polar and Garmin devices do work.

When a strap is connected, HUSH records the drop in resting heart rate across the settle phase. It is the one measurement here you can watch happen in real time, and it does not depend on anyone's sleep-staging algorithm.

Take it with you

Your nights, your trial, your results. Plain files, no lock-in.

How much to trust the numbers. Wrist and ring trackers are good at telling sleep from wake but only fair to moderate at naming the stage. One validation study put agreement with clinical sleep scoring in the range of a Cohen's kappa of about 0.2 to 0.5 across six devices, and concluded the better ones are suited to tracking prolonged changes rather than judging a single night. That is exactly why this feature randomises and needs a couple of weeks. One night proves nothing. Twenty randomised nights might. Chinoy et al. and related validation work; see also the 11-device multicentre comparison in JMIR mHealth 2023.

The newest and best-controlled study is not good news for all-night noise. A 2026 lab study put 25 adults through seven nights of sleep monitoring. Continuous pink noise did not protect sleep from traffic noise. It made sleep structure worse, and specifically shortened REM. Foam earplugs did work. The authors single out newborns and toddlers as the group to be most careful with, because REM matters most for a developing brain. Basner et al., SLEEP 2026;49(5):zsag001. University of Pennsylvania.

The broader picture was already uncertain. A systematic review of 38 studies graded the evidence that continuous noise improves sleep as very low quality, which sits oddly next to how widely these machines are used. Riedy, Smith, Rocha & Basner, Sleep Medicine Reviews 2021;55:101385.

If you are an adult, try earplugs first. In that same 2026 study foam earplugs did protect sleep, and continuous pink noise did not. A pair costs almost nothing and needs no app at all. Use this instead when earplugs are not an option: when you need to hear a child, when something in your ear is uncomfortable or unsafe, or when what you want is a wind down rather than a block out. None of that applies to babies, who should never be given earplugs.

What this app does about it. Every program here fades out instead of running until morning. Adaptive masking exists so the noise can stay off until something actually needs covering. The volume cap is on by default. None of that makes noise proven. It just means the app is built around the part of the evidence that is least likely to hurt: use it to get through the settling, then let the room go quiet.

Hearing, separately. All fourteen infant sound machines measured in a 2014 study exceeded 50 dBA at 30 cm on maximum volume, the hospital nursery reference level, and three went past 85 dBA, an occupational limit for adults over eight hours. Distance and volume are the whole fix. That is what the room meter is for. Hugh et al., Pediatrics 2014;133(4):677 to 681. AAP 2023 guidance: as far from the child as possible, as quiet as still works, and not left running longer than it needs to be.

Where it does look promising. Intermittent stimulation is a different thing from continuous noise. Short pink bursts timed into deep sleep raised slow-wave activity and memory scores in small studies, and an open-loop version raised the share of the night in N3. That is the one genuinely encouraging strand, and it is what the Deep sleep panel implements. Papalambros et al. 2017; Ngo et al. 2013; Schade et al. 2020.

Where these descriptions came from. The notes on each sound were written from published sources, meaning journal papers, Cochrane reviews, garden and instrument histories, by someone outside the traditions several of these sounds belong to. They aim to describe, not to speak for anyone. If something here is wrong, shallow, or misattributed, that is worth fixing and corrections are welcome.

None of this is medical advice and this is not a medical device. If a child's sleep, hearing or breathing worries you, that's a conversation for your paediatrician.

Volume cap

Limits maximum output. Leave this on for babies and toddlers.

Three things matter more than the app. Put the phone or speaker as far from the crib as the room allows, never on the rail. Keep it quiet enough to talk over in a normal voice from where the baby is. Use a timer instead of running it all night.