Pink Noise Generator

Pink noise is a random signal whose power falls off at −3 dB per octave from low to high frequency — equivalently, its power spectral density goes as 1/f. It is the broadband noise that contains equal power in every octave, which is why a spectrum analyser configured with a log-frequency axis (like the one below) shows it as a flat horizontal band. White noise, by contrast, has equal power per Hz and slopes upward by +3 dB/octave on the same display.

Why “equal power per octave” matters

Human hearing is roughly logarithmic in frequency — one octave (a 2× jump in Hz) is the same perceptual step whether it's 100→200 Hz or 5,000→10,000 Hz. Pink noise's flat-per-octave profile lines up with that perception, so it sounds balanced rather than “hissy” (white) or “rumbly” (brown). The same property is why pink, not white, is the standard test signal for audio measurement: it excites every octave with the same power so the room/speaker response is what the analyser is measuring, not the test signal's own bias.

Common uses

• Speaker / monitor calibration — flat-per-octave excitation paired with a real-time analyser shows the speaker's frequency response (deviations from flat) directly, without the +3 dB/oct correction needed with white noise.
• Room-mode measurement and RT60 — pink noise excites low frequencies more than white, surfacing room modes and reverberation patterns at musically-relevant levels.
• Sleep, focus, and infant soothing — the gentler high-frequency content (compared to white) is less fatiguing for long-session listening. Many sleep clinicians recommend pink over white for this reason.
• Music masking — pink's spectrum sits closer to the long-term average spectrum of speech and music, so it masks intrusions more naturally than the brighter white.

How this generator works

The audio is produced by feeding uniform white noise through Paul Kellet's six-pole IIR filter, a canonical cascade of one-pole low-pass sections whose coefficients are tuned to approximate the 1/f spectrum to within ±0.05 dB from ~9 Hz to ~22 kHz. The filter runs sample-by-sample inside the browser (no external library), and the output is written into a 2-second AudioBuffer that loops via a Web Audio AudioBufferSourceNode for indefinite playback at near-zero CPU.

About the “Compare colors” button

This plays three seconds each of white, pink, and brown noise in sequence so you can hear them back-to-back. White is the brightest (most upper-frequency content); pink sits in the middle; brown is deepest and rumbliest. Each colour uses its own generator: white via direct Math.random(), pink via the Kellet IIR cascade above, and brown via a leaky integrator on white (approximates −6 dB/octave).