Noise is the broadband test signal. Where a tone answers a question about one frequency, noise answers it about all of them at once.
The three colours
White has equal energy per hertz. Because each octave up contains twice as many hertz as the one below, white noise puts half its total energy in the top octave, which is why it sounds like a hiss rather than a balanced sound.
Pink falls at 3 dB per octave, which works out to equal energy per octave. This is the one that sounds even to the ear and the one measurement tools expect.
Brown falls at 6 dB per octave, produced here by integrating white noise with a leak to stop it drifting. It is the deepest of the three and the one usually reached for as a background sound.
Level and crest factor
The level control is peak amplitude, not RMS. Noise reaches its peak rarely and sits well below it most of the time, so a file peaking at -12 dBFS measures around -22 dBFS RMS.
When matching noise to a tone by level, match them by ear or with the loudness meter rather than by the number in the control.
Channels
Both channels are generated from separate states, so the stereo file is genuinely two signals. Summing it to mono partially cancels, which is exactly what should happen and is itself a useful test of a mono fold-down.
For a test that needs identical channels, generate a mono file and duplicate it in whatever plays it back.
Sample rate and bit depth
Noise fills the band up to half the sample rate. An 8 kHz file contains nothing above 4 kHz, which sounds muffled next to a 48 kHz one at the same colour.
24-bit gives a noise floor low enough that the file’s own dither is far below what is being tested.
Example: setting speaker levels
Pink noise, -12 dBFS, stereo, 30 seconds.
Play it through one speaker at a time and adjust until a meter at the listening position reads the same for each. Because pink noise has energy everywhere, one measurement covers the whole range instead of a tone that might land on a room mode.