How to Use a Noise Mixer: Complete Guide

Move one slider first

Switch on the white layer, drag its slider to 50, and listen before you touch anything else. What you have just done is set a gain node to the slider position divided by one hundred, times a fixed make-up factor of 1.25, so every sample of the generated signal is multiplied by 0.625 on its way to the output. The slider sets a gain multiplier, not a level in decibels. The default position is 80, where the multiplier is 1.0 and the signal passes through unchanged; 100 is the loudest the layer goes, about two decibels above that. Nothing in that chain knows how loud your headphones are, where your operating system volume sits, or what reaches your ear.

This is worth stating plainly because the slider carries a number from 0 to 100, and a number invites the reader to treat it as a measurement. It is not one. Two people with the same slider at 60 are listening at two different levels, and the same person switching from laptop speakers to closed-back headphones changes the level without touching the slider at all. Any instruction of the form set pink noise to thirty percent is advice about a multiplier, not about sound, and it does not transfer between setups.

Why halving the number does not halve the loudness

A gain node scales amplitude. Sound level in decibels is twenty times the base-ten logarithm of the amplitude ratio, so moving a slider from 100 to 50 halves the amplitude and drops the level by about six decibels. Moving from 50 to 25 drops it by another six. The slider is linear in amplitude and therefore compressed at the top and stretched at the bottom: the first ten points you give away take more level than the last ten, and the useful working range for a quiet background sits lower on the slider than most people expect.

One practical consequence is that small adjustments near the bottom of the travel are not small. Going from 4 to 2 is the same six-decibel step as going from 100 to 50. If a layer is already quiet and you want it slightly quieter, expect to move the slider very little.

The make-up factor exists because the generator holds every colour to the same root-mean-square level, about minus sixteen and a half decibels relative to full scale, so that changing colour changes the spectrum and not the loudness. That is a conservative level chosen to leave headroom, and the factor spends some of that headroom back. It is capped per colour at whatever would put the loudest sample of that buffer exactly at full scale, so a single layer at 100 cannot clip.

Five generators, five recipes

The mixer synthesises its sound in the browser rather than playing recordings, so what you hear is defined by arithmetic rather than by a microphone. Each of the five layers fills a ten-second buffer and loops it, with the last fifty milliseconds blended into the start at equal power so the join does not click or tick at a hearable interval.

White

Each sample is an independent random number spread evenly between minus one and one. That produces the flat power spectral density which defines white noise, equal power in every equally wide frequency band.

Pink

Pink is white noise fed through a weighted sum of seven first-order filters. The coefficients are Paul Kellett's refined pinking filter, published to the music-DSP archive, which approximates the required minus ten decibels per decade to within a twentieth of a decibel above about nine hertz at a 44.1 kilohertz sampling rate. Minus ten decibels per decade is the same slope as minus three decibels per octave, which is the defining property of pink noise.

Brown

Brown is white noise fed through a leaky integrator: each output sample is the previous one plus a small fraction of a new random value, then scaled back up. Integrating a flat spectrum tilts it by six decibels per octave, which is the brown slope, and it is also why the result wanders slowly rather than hissing.

Green

Green is white noise through two cascaded band-pass sections, both centred on 800 hertz with a Q of 0.5, which is the geometric middle of the 400 to 1600 hertz band and wide enough to sound like a band of noise rather than a pitch. Both the rumble and the hiss fall away at about twelve decibels per octave. Green has no standardised definition, so this is a choice rather than a specification: another site's green will not be the same filter.

Grey

Grey is white noise through three sections that roughly invert an equal-loudness contour: a low shelf lifting 200 hertz and below, a dip of eight decibels centred at 3.2 kilohertz where hearing is most sensitive, and a high shelf lifting 9 kilohertz and above. The intent is a sound that measures uneven and hears even. Grey has no standardised definition either, and a literal inverse of a published contour would need about forty decibels of sub-bass, which is not a usable sound.

What each control changes

Every control on the page acts on some stage of one signal path, and knowing which stage saves you from expecting a control to do a job that belongs to another one. The five layer sliders and their on/off buttons set the layer gains. The five-band equaliser sits after the layers are summed, so its filters shape every colour at once rather than one of them. The fade toggle and its two duration fields decide how fast a layer gain ramps up when a layer starts and down when the timer ends. The timer and the Pomodoro button ramp every layer to silence at zero.

Two consequences of that order are worth knowing. Because the equaliser is after the mix, there is no way to equalise one layer and not another; if you want less top end on white only, lower the white slider instead. Because the fade acts on the layer gains and not on the master, a fade-out leaves the equaliser and the visualiser exactly as they were, and switching a layer on during a fade-out ends that fade: the layers still fading ramp out over eighty milliseconds, and the new one comes up at its normal level rather than inheriting the fade or sitting under it.

Block diagram of the mixer's audio graph, left to right. Five source blocks — white from uniform random values, pink from a seven-pole filter, brown from an integrator, green from two band-passes and grey from three shelves — each described as a ten-second buffer crossfaded at the seam and scaled to the same loudness as the other four. Each source feeds its own layer gain block, whose value is the slider position over one hundred times 1.25, noted as 1.0 at the default slider of 80, and noted as the stage the fade ramps and the timer ramps down to silence. The five layer gains converge on a bus gain block labelled divided by the square root of N layers, which feeds a block of five peaking filters at 60, 250, 1000, 4000 and 16000 hertz, which feeds a master gain of 1.0, which feeds your output. A softer line taps the master gain for the visualizer. A note at the foot lists which control acts on which stage: the layer slider and on-off button act on the layer gain, the equalizer sliders on the five filters at plus or minus twelve decibels each, the fade toggle and durations on how fast a layer gain ramps up and down, and the timer and Pomodoro on ramping every layer gain to silence at zero, adding that no audio file is played but the alarm. Five generated colours Layer gain Bus gain Equalizer Master White — uniform random Pink — 7-pole filter Brown — integrator Green — two band-passes Grey — three shelves Each: a 10-second buffer, crossfaded at the seam, scaled to the same loudness as the other four slider/100 × 1.25 slider/100 × 1.25 slider/100 × 1.25 slider/100 × 1.25 slider/100 × 1.25 Default slider 80, so 1.0. This is the stage the fade ramps and the timer ramps down to silence. ÷ √N layers Five peaking filters 60, 250, 1k, 4k, 16k Hz Master, 1.0 Your output Visualizer tap Every control on the page acts on a stage above: Layer slider and On/Off — the layer gain. Equalizer sliders — the five filters, ±12 dB each. Fade toggle and durations — how fast a layer gain ramps up at the start and down at the end. Timer and Pomodoro — ramp every layer gain to silence at zero. No audio file is played but the alarm.
The mixer page's audio graph as implemented, with each control written against the stage it changes. Five generated colours each pass a layer gain, sum, and are divided by the square root of the number of active layers before the equaliser, the master gain and the output. The fade ramps the layer gains; the timer ramps them to silence.

With the fade toggle off, layers still ramp, over eighty milliseconds, which is long enough to prevent the click that an instant start or stop produces and short enough to feel immediate. The toggle chooses between that and the durations in the fields, defaulting to five seconds in and thirty out.

What the page does not do is remember anything. There is no saved preset, no remembered slider position and no history: reload the page and every layer is off with its slider back at the default of 80. Guides that tell you to name and recall your mixes are describing a feature this mixer does not have. The three scene buttons are the nearest equivalent, and they set fixed combinations rather than yours.

What layering actually buys you

Running two layers at once is not the same as turning one of them up. Two independent random signals add in power rather than in amplitude, so two layers at equal gain would land about three decibels above either one alone. The mixer divides the summed signal by the square root of the number of active layers, which is exactly that three decibels, so adding a layer changes the shape of the sound and holds the level where it was. What you gain by layering is spectrum, not loudness, and the volume control for loudness is the sliders.

That shape is the reason to layer at all. Brown noise alone has very little energy in the frequency range where speech sits, so it masks a rumbling fan well and a nearby conversation poorly. Adding a thin white layer puts energy back at the top of the range without giving up the low end. Pink sits between the two and needs less help. If nothing in your environment bothers you above a few hundred hertz, a second layer is adding level for nothing, and the single-layer mix is the better one.

Two jobs the mixer is not for

It will not make a loud room quiet. Masking replaces an irregular distraction with a constant one, which is a trade rather than a removal, and if the distraction is loud enough that covering it means an uncomfortable level, the sound is the wrong tool and distance or a closed door is the right one.

It will not reliably improve your sleep either. A systematic review of continuous noise as a sleep aid graded the evidence very low quality, with some studies finding disruption rather than benefit. That is a weaker claim than the one usually made for these tools, and it is the one the evidence supports.

Setting a level you can defend

No published source establishes an optimal listening level for background noise, so the honest rule is a ceiling and a procedure rather than a number on the slider. For the ceiling: the National Institute on Deafness and Other Communication Disorders reports that sound at or below 70 A-weighted decibels is unlikely to cause hearing loss even after long exposure, while long or repeated exposure at or above 85 dBA can cause it. Normal conversation is 60 to 70 dBA. All-day listening belongs well under the lower figure, and since the slider does not tell you where you are, the only way to know is to measure at your ear or to stay obviously quiet.

For the procedure: raise the layer until the thing you are trying to cover stops drawing your attention, then stop. There is a real cost to going further. A meta-analysis of sixty-five experiments on reading found that background noise, speech and music all carry a small but reliable penalty, with intelligible speech and lyrics the worst. The case for the mixer is not that noise helps you read; it is that a steady sound you choose can be less costly than an intermittent one you did not.

References

Frequently Asked Questions

How many layers should run at once?

As few as cover the sound that is bothering you. The mixer has five generated layers, and adding one does not raise the level: the mix is divided by the square root of the number of active layers, so two at equal gain sit where either sat alone. Add a second layer when the first leaves a gap in the spectrum, which usually means a low layer that does not reach speech frequencies.

What does the equalizer actually change?

The five sliders set the gain of five peaking filters, at 60, 250, 1000, 4000 and 16000 hertz, over plus or minus twelve decibels each. The filters sit after the layers are summed, so they shape the whole mix rather than one colour: boosting 60 hertz lifts the low end of every active layer at once. To change one layer only, move its slider instead.

Can a mix be saved and recalled later?

Not currently. Nothing on the page is stored, so reloading turns every layer off and returns each slider to its default of 80. The three scene buttons are the nearest thing to presets, and they apply fixed combinations rather than ones you define.

Does the timer stop the sound when it reaches zero?

Yes. The alarm rings and every active layer ramps to silence, over the fade-out duration if the fade toggle is on and over eighty milliseconds if it is not. The layers are left switched off, so starting again is a matter of switching the colours you want back on.

WhiteNoise.top

WhiteNoise.top builds free browser-based noise generators, mixers and breathing timers. Our articles explain the acoustics behind those tools and cite published research where it exists.