🎨

Noise Color Analyzer

This noise color analyzer records noise through your microphone and names its color — a practical noise color identifier for any broadband sound source. It fits a straight line to the time-averaged power spectrum plotted against log-frequency and reads the slope in decibels per octave: white ≈ 0, pink ≈ −3, brown/red ≈ −6, blue ≈ +3, violet ≈ +6, with grey treated as perceptually flat.

Spectral slope is a RELATIVE measurement, so this is one of the more trustworthy microphone tools — the color depends on the shape of the spectrum, not its absolute level, which means it is calibration-independent. Two honest caveats: (1) the fit is over the per-hertz spectral density, where pink noise truly slopes −3 dB/octave even though it looks “flat” on an equal-energy-per-octave display — don’t call pink flat without that qualifier; and (2) consumer mics roll off at the frequency extremes and cannot capture true infrasound (<20 Hz), so the slope is fitted over a mid band where your mic is most reliable. AGC / noise-suppression are requested off automatically (a reading is meaningless if your OS keeps them on). Nothing is recorded or uploaded.

Microphone

Fit band (Hz)

The slope is fitted over this range. The default avoids deep bass (mic roll-off / room rumble) and the very top octave (mic roll-off).

Idle — press Start, then play steady noise.
Slope
Color
Fit R²
Confidence
Best-match color

Optional: calibrate level for a dB SPL estimate

Captured level: . This offset is shared across every noise tool on the site and does not affect the color/slope result, which is level-independent.

The noise colors

White ≈ 0 dB/octEqual power per hertz. Bright, hissy. Masking and testing.
Pink ≈ −3 dB/octEqual energy per octave (so it looks flat per-octave but slopes −3 in density). Balanced; used for calibration and sleep.
Brown / red ≈ −6 dB/octEnergy ∝ 1/f². Deep, rumbly. Relaxation, low-frequency tests.
Blue / azure ≈ +3 dB/octMirror of pink. Thin, crisp. Dithering, high-frequency masking.
Violet / purple ≈ +6 dB/octDerivative of white. The brightest, most treble-heavy color.
Grey perceptually flatShaped by a loudness curve so every band sounds equally loud — not a single straight-line slope.

Share or embed this tool

Free to use on your own website — WordPress, Wix, or any platform. Paste one line and it works instantly, resizing to fit.


How It Works

“Color” is just a name for how a noise’s energy is distributed across frequency — a concept that originates in the analogy between the visible spectrum and the power spectral density of a random signal. The power spectrum slope — how steeply the density rises or falls across octaves — is what defines each color. The tool captures your microphone, runs a large FFT (16,384 points), and time-averages the magnitude spectrum so a steady noise settles into a stable curve. It then plots that curve in decibels versus log₂(frequency) — i.e. against octaves — and fits a straight line by least squares. The line’s gradient, in dB per octave, is the spectral slope, and the nearest canonical slope names the color: 0 is white, −3 is pink, −6 is brown/red, +3 is blue, +6 is violet.

The single most important subtlety is the axis. The analyser returns a per-bin magnitude, and because FFT bins are evenly spaced in hertz, each bin represents a fixed slice of bandwidth — so the curve we fit is the per-hertz spectral density. On that axis, pink noise genuinely slopes down at −3 dB/octave. Pink is famous for having equal energy per octave, which makes it look flat on an equal-energy-per-octave (real-time-analyzer style) display — but that flat appearance and the −3 dB/octave density slope are two views of the same signal, not a contradiction. This tool reports the density slope, so it will read pink as −3, not as flat. White noise is flat in density (0 dB/oct) and rises +3 dB per octave on the equal-energy display, which is why it sounds so bright. Pink noise — also known as 1/f noise in the electronics and physics literature, because its power is inversely proportional to frequency — is the most common color found in natural sounds, including rainfall, heartbeats, and many HVAC systems.

Because the result is a slope — a relative comparison of high frequencies to low — the absolute level is irrelevant. That makes the color classification calibration-independent: an uncalibrated consumer mic that cannot tell you the true dB SPL can still tell you the shape of the spectrum reliably, which is why this is one of the more trustworthy microphone tools on the site. The honest limits are at the edges: phone and laptop mics roll off in the deep bass and the very top octave and cannot capture true infrasound, so the fit is restricted to a mid band (you can adjust it) where the mic is dependable, and a steeply tilted mic response can still bias an extreme reading. The value tells you how straight a tilted spectrum really is — a high R² means a clean single-slope color. A flat (white) spectrum is the exception: it has no trend for the line to explain, so its R² is near zero even when it is perfectly clean, and the tool instead judges the near-flat case by how tightly the spectrum scatters around flat — tight scatter reads as white, while a bumpy, irregular near-flat spectrum is reported as grey (perceptually flat / unstructured). Acoustic researchers and audio engineers sometimes call a grey or near-flat broadband signal wideband noise, which is a useful reminder that “broadband” does not imply a specific spectral shape. To verify the color of a generated reference signal, you can run it through the noise comparison tool to quantify how the spectrum changes when you switch noise colors.

Frequently Asked Questions

Is pink noise flat or sloped? My other app shows it flat.
Both are correct — they are different axes. Pink noise has equal energy per octave, so on an equal-energy-per-octave (octave-band / RTA) display it looks flat. But in per-hertz spectral density it slopes down at −3 dB per octave, because each higher octave spreads the same energy over twice the bandwidth. This tool fits the per-hertz density, so it correctly reports pink as −3 dB/oct, not flat.
Why is this more trustworthy than a decibel meter in a browser?
Because it measures the spectral slope — the shape of the spectrum — rather than an absolute level. Slope is a relative comparison of high to low frequencies, so it does not depend on your microphone’s unknown sensitivity or on any calibration. A browser mic cannot give a true dB SPL, but it can reliably show whether a spectrum rises or falls and how steeply, which is exactly what naming a noise color needs.
Do I need to calibrate the microphone?
Not for the color result — the slope is level-independent, so no calibration is required. The optional calibration box only adds an approximate dB SPL estimate of the captured level: read the level on a real sound-level meter and enter it, and the offset is saved and shared across every noise tool on the site. It never changes the color or slope, and the SPL figure remains an uncalibrated estimate, not a certified measurement.
Why must noise suppression and automatic gain be off?
Automatic gain control, noise suppression and echo cancellation all reshape the spectrum — they pull down steady broadband noise and tilt the frequency balance, which is exactly the shape this tool measures. The reading would be meaningless with them on, so the tool requests raw microphone audio with all three switched off. If your operating system still applies processing at the driver level, look for a “raw” or “disable enhancements” option.
Can it measure infrasound or very deep sub-bass?
No. Consumer phone and laptop microphones roll off sharply in the deep bass and cannot capture true infrasound below about 20 Hz, and they also roll off in the very top octave. That is why the slope is fitted over a mid band (80 Hz to 10 kHz by default, adjustable) where the mic is dependable. Reading the slope right down to a few hertz would mostly measure the microphone, not the noise.
What does a low R² or “grey” result mean?
R² measures how well a sloped straight line fits the averaged spectrum, so it is most meaningful for the tilted colors: a high R² on a clearly rising or falling spectrum means one clean slope (a definite color), while a low R² on a tilted spectrum means it is irregular — music, speech, a tonal hum, or a mix — so the label is only the nearest match. White is the exception: a truly flat spectrum has no trend for the line to explain, so its R² is near zero even when it is perfectly clean. For the near-flat case the tool therefore judges white vs grey by how much the spectrum scatters around flat, not by R² — tight scatter is reported as white, while a near-flat but irregular (bumpy) spectrum is labelled grey (perceptually flat / unstructured).
Is my audio recorded or uploaded?
No. The microphone signal is analyzed live in your browser to compute the spectrum and slope, and is never recorded, saved or transmitted. The microphone is connected only to an analyser node — never to your speakers — and all tracks are stopped and the audio context closed when you press Stop or close the tab.
My slope is between two canonical values — what does that mean?
Real-world noise rarely lands exactly on a canonical slope. A reading of −4.5 dB/oct, for instance, falls between pink (−3) and brown/red (−6), and the tool will name the closest match with a lower confidence score. This is normal: HVAC ducts, ventilation fans, and many natural sound sources have power spectral density slopes that fall partway between the idealized colors. The slope value itself is the meaningful measurement; the color name is just a convenient label for the nearest archetype.
How is grey noise different from white noise if both appear flat?
White noise has a mathematically flat power spectral density — equal power per hertz — so it sounds bright and hissy because the human ear is most sensitive in the 2–5 kHz range. Grey noise is deliberately shaped by an inverse equal-loudness (loudness) curve so that every frequency band sounds equally loud to the average listener; it is perceptually flat rather than spectrally flat. Grey does not have a single definable dB/octave slope, which is why this tool classifies it separately from the tilted colors. To generate a reference grey signal you would need a noise generator with a built-in loudness-curve filter.
Can I use this to check the color of a noise generator or sleep-sound app?
Yes — that is one of the most practical uses. Play the generator through a speaker, hold the microphone a short distance away in a quiet room, and run the analyzer. The fitted slope tells you whether the generator is actually producing the advertised color. Bear in mind that speaker coloration and room acoustics will bias the reading at the frequency extremes, so focus on mid-band agreement rather than a perfect match to the expected dB/oct value. To cross-check the spectral shape visually, use the noise floor analyzer to see the octave-band breakdown alongside this tool’s slope reading.
What noise color is typical for HVAC, fans, and environmental background noise?
Most real-world broadband noise sources — ventilation fans, air conditioning units, and general indoor ambient noise — produce spectra closer to pink or brown/red (slopes between −3 and −6 dB/oct) than to white. This is because low-frequency energy tends to dominate in mechanical systems and in large rooms where acoustic modes are closely spaced. Outdoor environmental noise is similarly tilted toward the lower frequencies. A truly white broadband source (flat per-hertz density) is much rarer in practice and is more commonly encountered in electronic test equipment than in natural or architectural acoustics.