How this noise is generated

The signal chain

Each colour runs through the same five stages. The only thing that differs between colours is the source algorithm and one filter.

source → colour filter → 35 Hz high-pass → gain → limiter → output

1. Source. An 8-second stereo buffer, generated fresh each session with independent samples per channel. Eight seconds rather than the more common two or four: brown noise wanders slowly, and on a short buffer that wander becomes an audible repeating shape.
2. Conditioning. DC offset is measured and subtracted, then the buffer is normalised to a 0.85 peak so every colour arrives at the gain stage at the same level with headroom left over. The tail is crossfaded onto the head across 60 ms using square-root weights, which keeps RMS constant through the seam because the two segments are uncorrelated. Without that crossfade, brown noise jumps roughly 0.52 in amplitude at the loop point — a hard click, every eight seconds, forever.
3. Colour filter. One biquad per colour (table below). Peaking and shelf filters get a +6 dB gain, because at the default 0 dB they do nothing and the colours would be indistinguishable from their base noise.
4. 35 Hz high-pass at Q 0.707. Removes content no speaker can reproduce but every speaker still tries to move for.
5. Gain and limiter. Gain is the slider value squared. A DynamicsCompressor at −3 dB threshold catches any peak the filter stage adds.

Filter settings per colour

These are the actual values this site runs, not a general description of what the colours mean. Published because no other free noise generator states theirs, which makes it impossible to tell whether two sites playing “green noise” are playing anything like the same thing.

Source algorithms: white — Uniform random samples, no filter. Flat power across the spectrum. pink — Paul Kellet 7-pole approximation, −3 dB/octave. brown — Leaky integrator, last = (last + 0.02w) / 1.02, −6 dB/octave.

What this site will not tell you

Colour noise is genuinely useful for masking and, for a lot of people, for focus and sleep onset. The evidence for anything beyond that is thinner than the wellness internet suggests, and this site tries to say so on the pages where it matters:

• Roughly a third of people in noise-and-attention studies perform worse with noise than without it. If you try brown noise for two weeks and nothing improves, that is a real result, not a failure of technique.
• The pink-noise-and-deep-sleep research that gets cited everywhere mostly used noise pulses timed to slow-wave activity in a sleep lab with electrodes attached. That is not the same as pink noise from a speaker, and the transfer is not established.
• Volume matters more than colour. 45–60 dB measured at your pillow, not at the device. Brown noise in particular feels 15–20 dB gentler than it measures, because the ear is least sensitive where brown noise puts its energy — which is why people play it far louder than they intend to.

Who runs this

Color Noises is an independent site, built and maintained by one person since 2026. It exists because the free noise generators available at the time all played short looped files, and the loop was audible. Reach us at hello@colornoises.com — corrections to anything factual on the site are welcome and get published with the change.

How we handle money and data

Gear recommendations carry affiliate links, and the site runs display advertising. No advertising runs inside the player itself. Recommendations are not ordered by commission rate, and where a product is only worth buying for a narrow use case the page says so. Full terms are on the affiliate disclosure page.

Your listening stats — time, sessions, streak, volume per colour, timer preference — live in your browser under the localStorage key colornoises:v1. They are never transmitted. There are no accounts. See the privacy policy.