White, pink and brown noise

What is noise in audio?

Noise is a signal where every sample is random. A sine wave repeats the same shape over and over, so it has one pitch; noise never repeats, so it has no pitch at all and spreads its energy across the whole frequency range instead. You know the sound already: the hiss of a detuned radio, steam from a kettle, wind, waves on a beach.

In mixing, noise usually means something unwanted, like tape hiss or a buzz from a cable. This page is about noise as a sound you choose. Synths have a noise oscillator alongside their regular waveforms (the oscillator waveforms lesson covers those), engineers use noise to test speakers and rooms, and the snares and claps in a lot of electronic drum kits get their rattle from filtered noise.

The colours describe the shape of that spread. Every colour is just as random as the others; what changes is how much energy sits low compared with high. On a spectrum that shows up as a straight line tilted by a set number of decibels per octave, and the tilt is the whole difference between them. The Slope knob in the lab does exactly that, sliding the same noise from brown to violet.

Per hertz or per octave: the idea behind every colour

Hearing works in octaves. The jump from 100 Hz to 200 Hz sounds the same size as the jump from 5 kHz to 10 kHz, even though the first spans 100 Hz and the second spans 5,000. Every octave is twice as wide in hertz as the one below it.

That doubling is where the colours come from. Divide the range people hear, roughly 20 Hz to 20 kHz, into octaves and you get about ten of them, starting at 20 to 40 Hz and ending at 10 to 20 kHz. The bottom octave is 20 Hz wide. The top one is 10,000 Hz wide, 500 times as many hertz.

White noise puts the same energy into every hertz. So the top octave, holding half of all the hertz between 0 and 20 kHz, holds half of all the energy too, and each octave going down holds half as much as the one above it. Halving the power is a 3 dB drop, which is why the octave bars in the lab climb about 3 dB per octave for white noise even though its spectrum is flat.

Pink noise fixes that by letting the level in each hertz fall 3 dB per octave. The octave above has twice as many hertz at half the power each, so it ends up with exactly the same energy as the octave below, and with ten octaves sharing equally, each gets about 10% of the total. Brown noise falls 6 dB per octave per hertz, twice as steep, so its octaves lose 3 dB each going up.

When people say a colour falls "3 dB per octave", they nearly always mean the level per hertz, the first number column above. Keep that in mind when you read an analyser, because some show one and some show the other, which is covered further down.

White noise

White noise has a flat spectrum: every hertz gets the same energy on average. Digitally it's the easiest noise to make, because a stream of independent random numbers is white noise already, and the noise oscillator on a synth is white unless it says otherwise.

Half of its energy between 20 Hz and 20 kHz sits in the top octave, 10 to 20 kHz, so white noise sounds bright and hissy and gets tiring to listen to. In the lab, White draws a level line on the spectrum and octave bars that climb from left to right, and the coach measures the 8 kHz octave holding 21 dB more energy than the 63 Hz one, seven octaves at 3 dB each.

That brightness is what makes it useful in sound design. Short bursts of high-passed white noise make hi-hats and shakers. A burst shaped by a fast envelope and mixed under a tone gives you a snare; on the Roland TR-808 the snare's Snappy knob sets how much noise goes in. Long risers and sweeps before a drop are usually white noise through a low-pass filter whose cutoff climbs over a few bars, often with some resonance so the sweep whistles.

White noise also turns up where you can't hear it. The plain dither added when a master is reduced to 16 bits is flat noise at about −96 dBFS, which the dither lesson explains.

Pink noise

Pink noise loses 3 dB per octave in each hertz, so every octave holds the same energy. You'll also see it written as 1/f noise, because the power in each hertz is in proportion to one over the frequency: double the frequency and the power per hertz halves.

To most people it sounds even, with no part of the range sticking out. That's the octave idea again. Your ears split the range into octaves and pink noise gives each of them the same share, so it sounds fuller and darker than white noise, more like rain on a roof than steam.

Pink noise is the studio's reference noise. Finished mixes, measured over a whole song, fall away towards the top of the spectrum, so they sit much closer to pink than to white, and pink noise through a speaker excites every octave of the room equally. Room measurement software runs on it, and so do the octave-band analysers used to tune live sound systems. Cinema mixing rooms, for example, are calibrated by playing pink noise at −20 dBFS RMS through each speaker in turn and setting each one to the same reading on an SPL meter at the mix position.

Brown noise

Brown noise falls 6 dB per octave in each hertz. Its name comes from Brownian motion, the random drift of tiny particles in water, which the botanist Robert Brown watched under a microscope in 1827. Brown noise moves the same way, because you can make it by adding up white noise: each sample is the previous one plus a small random step, so the wave wanders instead of jumping, which is exactly what the close-up in the lab shows. It's also called red noise.

With that much energy in the low end, brown noise sounds like a rumble, closer to distant thunder or the cabin of a plane than to hiss. In the lab, pick Brown and the coach shows most of the power sitting below 200 Hz and almost none above 5 kHz.

In music it's good for wind and sea sounds, and as a low bed under ambient pads. It's also the one to be careful with, because it fills the bass. A brown noise layer that sounds quiet on laptop speakers can be eating headroom and muddying the kick on a system with a sub.

Blue, violet and grey noise

Blue noise rises 3 dB per octave, the mirror of pink, and violet noise rises 6 dB per octave, the mirror of brown. You get violet noise by taking the difference between each white noise sample and the one before it, which strips out the low end. Both sound thin and sharp, all hiss with nothing under it.

You'll rarely reach for them in a track. They turn up in dither instead: noise-shaped dither pushes its noise towards the top of the spectrum, where your ears are less sensitive, so its spectrum climbs at the top end like blue or violet noise. Image software and printers use blue noise patterns for a similar reason, to hide grain where your eyes notice it least.

Grey noise is the odd one out. It's shaped by an equal-loudness curve so that a typical listener hears every frequency as equally loud, which means extra energy at the very bottom and top where hearing is weakest, and less around 2 to 5 kHz where it's sharpest. The curve changes with playback level, so grey noise only sounds even at the level it was made for, and the lab sticks to the five colours that are straight lines.

Comparing colours at the same loudness

Play white and brown noise at the same peak level and one sounds far louder than the other, which makes it hard to hear what the colour itself does. The lab measures each colour in LUFS, the loudness unit streaming services use, and plays them all at the level the Level knob sets, so switching colour changes the tone and leaves the loudness alone.

The coach shows what that does to the ordinary meters. At −26 LUFS the lab measures white noise at −29.1 dBFS RMS, pink at −26.6 and brown at −20.7, so brown reads more than 8 dB higher than white at the same loudness. That's because LUFS uses K-weighting, which counts deep bass for less and the treble a little more, roughly the way your ears do. A meter without that weighting overstates how loud brown noise is and understates white.

Step 5 sets up the comparison. Press A/B and the lab flips to pink noise at the same LUFS, or to white if pink is already playing, and the faint line on the spectrum shows the one you're flipping to.

Reading noise on a spectrum analyser

Put pink noise through the analyser in your DAW and you might see a flat line, or a line that falls from left to right. Both are right, because the two kinds of analyser are measuring different things.

An FFT analyser measures the level in each narrow slice of hertz, like the per-hertz spectrum in the lab. On one of those, white noise is flat and pink noise falls 3 dB per octave. An octave-band or third-octave analyser, the kind with a row of bars, adds up everything inside each band instead, so pink noise comes out flat and white noise climbs 3 dB per octave. The lab's octave bars work that way.

Plenty of analysers sit in between: they measure per hertz and then tilt the display so pink noise reads flat, because a tilted display makes a mix look level. Voxengo SPAN has a slope setting that does this, and the spectrum in the EQ lessons on this site is tilted 3 dB per octave for the same reason. If you're not sure which kind you're looking at, play pink noise through it and see which way the line goes.

Noise under a mix: hearing where each colour sits

Louder sounds hide quieter ones at nearby frequencies. Engineers call that masking, and noise is a clear way to hear it, because it puts a known amount of energy in every band. Pick Noise and loop in the lab and a loop plays at −20 LUFS under the noise. The pale marks on the octave bars are the loop's level in each octave, and wherever a coloured bar rises above its mark, the noise is louder than the music there on average and covers it up.

White noise beats the full mix only in the top octaves, so the hi-hats vanish while the bass is untouched. Brown noise at the same loudness covers the bass and kick and leaves the hats alone. Try it with the vocal too. At the same settings, white noise covers the 8 kHz octave, where the breaths and s sounds are, and the thin 125 Hz octave under the voice. Brown noise covers only that 125 Hz octave, so the words stay clear.

How to make a white noise riser

The rising whoosh before a drop is the most common use of noise in electronic music. Any synth with a noise oscillator and a filter can make one, and so can a white noise sample with your DAW's filter on it.

1. Start with white noiseSwitch off the synth's other oscillators and turn up the noise oscillator, or load a white noise sample, and write one long note across the build, often 8 or 16 bars.
2. Add a low-pass filterSet the cutoff low, around 200 to 400 Hz, so the riser starts as a dull rush. A band-pass filter works too and sounds thinner.
3. Automate the cutoff upwardsDraw the cutoff rising to 10 kHz or higher over the build. A curve that rises slowly at first and faster near the end sounds more urgent than a straight line.
4. Add a little resonanceA moderate resonance boost puts a whistling edge on the sweep. Too much and it turns into a sine-like whine.
5. Bring the volume up with itAutomate the level so the riser starts quiet and finishes loud, then cut it dead at the drop, or let a long reverb on it carry over the downbeat.
6. Clean up the low endHigh-pass the riser at 100 to 200 Hz so it never fights the kick and bass, which matters more as the cutoff opens.

The same building blocks make drums. Give the noise a short envelope, with no attack and a decay of 50 to 150 ms, then high-pass it around 6 to 8 kHz and you have a closed hi-hat. Lower the high-pass to 1 kHz or so, lengthen the decay and layer it under a snare, and the snare gets its rattle.

Using pink noise to check a mix and your speakers

Pink noise makes a quick check on a mix balance. Play it at a fixed level well under the mix and listen for which parts stay clear and which disappear into it. The parts that vanish first are the quietest in their range, so if the vocal goes under before the guitars do, the vocal may be lower in the balance than you thought. Step 8 sets up pink noise 12 dB under the full mix to try it.

It's also how you match the levels of two speakers. Play pink noise through one speaker at a time, measure each with an SPL meter or a phone app from the same listening position, and adjust until they read the same. Pink noise works better than a sine for this. Reflections make a single sine read several dB differently a short step away, while pink noise covers every frequency at once, so those peaks and dips average out.

Keep it quiet while you work. Continuous noise at a high level is tiring and can damage your hearing over a long session, and the hearing lesson covers safe listening levels.

Common mistakes with noise

Comparing colours at different levels

Whichever noise is louder will sound like the better one, or the more annoying one, and you learn nothing about the colour. Match the loudness first, as the lab does, then compare.

Reading an analyser the wrong way

Pink noise that looks flat on one analyser and sloped on another is normal. Check whether the analyser shows per hertz, per octave or a tilted view before you decide something is wrong with your noise, or your mix.

Leaving the low end in noise layers

Brown and pink noise carry a lot of bass. Under a mix that low end muddies the kick and bass and eats headroom, so high-pass noise layers unless the rumble is the point.

Using white noise when you wanted something softer

A white noise bed for ambience gets harsh quickly. Pink or brown noise, or white noise through a gentle low-pass, sits under music much more comfortably.

Training your ears to hear noise colours

The quiz under the lab plays white, pink or brown noise at the same loudness with the screens covered and asks which it is. Listen for how much hiss is on top and how much rumble is underneath. Eight rounds take a couple of minutes, and your best score is saved in your browser.

Questions people ask

Keep going