Brown Noise Generator

The sound below is being built as you listen, one sample at a time, in your own browser. There is no file, no stream, and no loop to catch.

What Is Actually Happening When You Press Play

Nothing was downloaded to make that sound. No audio file was fetched, nothing is streaming, and if you disconnect right now it will keep going. What arrived when this page loaded was roughly nine kilobytes of instructions for how to build brown noise, and your browser has been building it ever since — 48,000 fresh samples every second, on a dedicated audio thread, for as long as you leave it running.

That is the difference between a generator and a player, and it is the reason this page exists separately from the rest of the site. Most things calling themselves a brown noise generator are an audio file on a loop. This is a generator in the literal sense: a random number source, a filter, and arithmetic running in real time.

If you want to know what brown noise is — how it differs from white and pink, what it masks well, and what the research actually supports — that is a different question and it has its own page. This one is about the tool.

A Real Generator, or a Looped File in Disguise?

This is the question worth asking of any noise site, so here is the honest answer for this one, including the part that is a caveat rather than a boast.

On the main path there is no loop point at all. Modern browsers provide an AudioWorklet: a place to run your own audio code on the real-time audio thread. The generator runs there. Each block of samples is produced on demand and then discarded — no buffer is being replayed, because no buffer exists. Leave it on for eight hours and no second of it repeats any earlier second. The "can you hear the loop?" question does not get a good answer here so much as it stops applying.

On old browsers there is a fallback, and it does loop. If AudioWorklet is missing or fails to load, the page renders 30 seconds of noise in advance and plays it in a circle. It is built to be seamless — the random sequence is rewound to its exact starting state so the end of the buffer joins the beginning without a click — but it is a loop, and a page that told you otherwise would be lying about its own fallback. In practice this affects browsers older than about 2018. If yours is current, you are on the generator.

What the generator is made of, briefly. A splitmix32 pseudo-random source feeds an Irwin-Hall sum of four samples, which produces near-Gaussian white noise rather than the uniform kind — natural noise sources are Gaussian, and the amplitude distribution is audibly smoother for it. That white noise then passes through a one-pole low-pass filter cornered at 12 Hz, which is exactly −6 dB per octave above its corner, and a 6 Hz high-pass that removes sub-audible energy which would otherwise eat headroom and rattle phone speakers without contributing anything you can hear.

And it was checked, not assumed. The brown noise page sets out the measurement method in full; the short version is that this generator lands on −5.98 dB per octave against a −6 target, holding within a third of a decibel of a straight line across the audible band. Two things a spectrum plot will not show you. Every colour here arrives at the same loudness — −20 LUFS, K-weighted to ITU-R BS.1770-4 — so changing colour never means reaching for the volume knob. And the two channels run on separately seeded random sequences, so what reaches your left ear is not a duplicate of what reaches your right.

If You Came Looking for a Download

Some people arrive here searching for a brown noise file to download. There isn't one, and that is a deliberate choice rather than an oversight — so here is the reasoning, and what to do instead.

This site used to serve a 160 kilobyte MP3. It was replaced by the generator, and the reason is the one above: a downloaded file is by definition a fixed length, so overnight it must repeat. Ten hours of sleep from a two-minute file is three hundred repetitions, and the ear is very good at learning a pattern it has heard three hundred times. A generator has no such limit.

For a single session, this page is the download. Once it has loaded, the generator runs entirely on your device. Turn off wifi and it continues, because there is nothing left to fetch. That is the practical thing most people wanted a file for.

For overnight or offline use, install the app instead. A browser tab cannot legitimately keep playing once your phone locks — the operating system suspends it, and generating audio is exactly the kind of work a suspended tab stops doing. That is not a limitation of this page that a better page could fix; it is a rule browsers enforce on every website, and it is the reason sleep apps are installed rather than bookmarked. The Android app plays with the screen off, works with no connection at all, and has a sleep timer.

There is no email wall here and no file to unlock. If you want the sound now, it is above. If you want it overnight, it needs to be an app, and that is true of every noise website you will find.

How to Catch a Looping Generator in About Two Minutes

This is worth knowing generally, because "generator" is an unpoliced word and most sites using it are playing a file. None of these tests need any equipment beyond headphones, and all of them work on this page as readily as on a competitor's.

Listen for a landmark. Real noise occasionally throws up something faintly distinctive — a swell, a moment that sits lower than the rest. Fix on one, then wait. If it comes back at a tidy interval, you have found the loop length. Loops are usually 10, 30 or 60 seconds, so two minutes of attention is enough to catch almost all of them.

Watch the network, not the page. Open your browser's developer tools, go to the Network tab, and reload. A generator fetches a small script and nothing further. A player fetches an audio file, and you will see its name and its size sitting there in the list. This is the test that cannot be argued with.

Disconnect after it starts. Turn off wifi thirty seconds in. Anything streaming will run out and stop. A downloaded file will keep playing, and so will a generator — so this test alone does not distinguish those two, which is why it is second rather than first.

Be suspicious of a download button. Not because downloads are bad, but because a genuine generator has nothing to put behind one. If a site offers both, the file is the product and the generator is decoration.

Run the first two on this page if you like. The network tab will show you a script and no audio, and the landmark test will not find anything, because there is nothing to find.

Getting Something Useful Out of It

Headphones, or a speaker with a real bass driver. This matters more for brown noise than for any other colour, because most of its energy sits below what a phone or laptop speaker can physically move. Through a phone you are hearing the thin top of it, and turning the volume up only makes the thin part louder. People who conclude brown noise is not for them have very often only heard it reproduced badly.

Set the level against the noise you are covering, not against silence. The sound should sit just above the thing you are trying to mask, and no higher. Louder is not more effective — past a point you have simply swapped one intrusion for another, and you are also spending the night at a volume that is not doing your ears any favours.

Give it five minutes. The first thirty seconds of any noise sounds like an imposition. Whether it works is a question about minute four, when you have stopped listening to it and it has started doing its job.

Sessions run 30 minutes. The counter on the player is the free allowance. It is one budget for the whole site, so opening a second page does not hand you another half hour. The last minute is a slow fade rather than a switch being thrown, on the theory that an abrupt silence is its own kind of alarm.

Try the other colours before settling. Which one suits you depends on what you are covering, and it is not knowable in advance: brown for traffic rumble and machinery, white or green when the intrusion is speech through a wall, pink as a middle ground. Each runs on this same generator, so the comparison is fair.

What This Page Cannot Do

It will not survive your phone locking. Worth repeating, because it is the commonest disappointment with any browser-based noise tool: lock the screen or move to another app and the operating system suspends the tab, which stops the generator mid-sample. No website is exempt. So this works if you are drifting off with the phone awake and charging beside you, and it will not carry you to morning.

It needs JavaScript and the Web Audio API. There is no version of a browser-based generator that works with scripting disabled, because the arithmetic is the sound.

It will not mix with rain or thunder. One colour at a time here. Layering is an app feature, mostly because decoding several streams in a browser tab overnight is not a plan.

It is not a treatment for anything. Tinnitus and ADHD each have a page here, and each lands on a thinner evidence base than this whole category likes to imply. Insomnia is the same story. If sleep has been bad for weeks, the useful next step is a clinician, and no generator or app alters that.

If You Only Read One Paragraph

This is a real generator: brown noise built live in your browser at a measured −5.98 dB per octave, normalised to −20 LUFS, with independent left and right channels and no loop point on any current browser. It is free, it needs no account, and there is nothing to download — turn off your connection after it starts and it keeps running. Use headphones, set the level just above whatever you are masking, and give it five minutes before judging. If you want it running overnight with the screen off, that has to be an installed app, and the Android one is free.

Frequently Asked Questions

Related pages

• Brown Noise — The deepest colour, measured at −5.98 dB per octave — best for traffic rumble and long overnight sessions.
• Brown Noise App — Which brown noise app to install, whether it is really free, and the five things an installed app does that a browser tab is not allowed to.
• Brown Noise for Sleep — The free web player, plus the full guide to brown noise for sleep, focus, and tinnitus.
• White Noise — The flat reference colour, measured at 0.00 dB per octave — the best of the five at making speech unintelligible.
• Pink Noise — Equal energy in every octave, measured at −3.07 dB per octave — the natural-sounding default and the audio industry's calibration signal.
• Green Noise — The mid-frequency colour — a pink base with a broad 500 Hz emphasis, and the only one of the five with no standard definition.
• Blue Noise — The bright end of the spectrum at +3.08 dB per octave — an engineering signal that is genuinely useful against high-pitched noise, and a poor choice for sleep.
• Brown vs White Noise — The widest gap on the site — 5.98 dB per octave, about 38 dB apart by 8 kHz — and what each colour can and cannot cover as a result.
• Brown Noise for Focus — Why concentration is a masking problem measured in speech intelligibility, and why brown noise beats every colour for rumble but loses to white for voices.
• Brown Noise for Tinnitus — Masking versus habituation, why the pitch of your tinnitus decides the colour, and where sound therapy actually sits in the 2014 clinical guideline.