noisemoon › blog › youtube white noise isn't white

Most YouTube "white noise" isn't white. We measured 19 of the most popular videos.

Published 2026-05-05 · 12 min read · noisemoon.com/blog

Spectrally pure white noise has a flat power spectrum: equal energy at every frequency, all the way to 20 kHz. The 19 most popular YouTube videos labeled "white noise" averaged a spectral slope of −9.46 dB/oct, where pure white is 0. Most aren't white. Some are darker than proper brown noise. Six of nineteen are looped. Six of nineteen are likely mono masquerading as stereo.

The TL;DR

We downloaded and spectrally analysed 62 of the most-played YouTube videos in the white / pink / brown noise categories. Across the board, the videos drift dramatically darker than the textbook spec for the color they claim. Here are the headline numbers:

Read that table closely. The average "white noise" video on YouTube has a spectral slope steeper than proper brown noise should be (−9.46 vs the −6.0 brown ideal). Pink videos are about twice as steep as the spec. Brown videos are about twice as steep as the spec. None of these are small deviations.

If you've been listening to "white noise" on YouTube and feeling like it's a bit dark and rumbly, your ears are right. It is.

What white, pink, and brown are supposed to be

The "color" of noise refers to its power spectral density — how much energy sits at each frequency. The names come from analogy with light:

• White noise is flat. Equal power at every frequency from roughly 20 Hz to 20 kHz. Slope of 0 dB/oct. Sounds like static or a distant rush.
• Pink noise rolls off at −3 dB/oct — halving in power for each doubling of frequency. Sounds like steady rain or a distant waterfall.
• Brown (or red) noise rolls off at −6 dB/oct. Deeper still. Sounds like surf, a distant freight train, or a low rumble.

These are mathematical definitions. A signal generator producing them is exact: feed in the right transfer function, get exact white / pink / brown out. Audio engineers use them as reference signals because they're well-defined. Sleep researchers cite them because the spectral content predicts the masking behaviour.

Almost no YouTube video labeled with these names actually matches the spec.

The survey

62 videos. The most-played in each color, sampled in late April 2026:

• 19 videos labeled "white noise" (or close variants like "pure white noise")
• 22 videos labeled "pink noise"
• 21 videos labeled "brown noise"

For each video, we extracted a 60-second segment from the middle (avoiding intro fades), down-mixed for the slope analysis, and ran a spectral fit between 80 Hz and 8 kHz — the band where the noise color is most consistently defined. We also kept a separate stereo path to measure left-right correlation, and ran a self-similarity check across the segment to detect loop seams. Methodology details are at the end of this post.

The white-noise result, in detail

Pure white noise has its power evenly distributed across the audible spectrum. So when you look at a third-octave plot, every band is roughly the same height — and a least-squares fit to "dB versus log-frequency" gives you a flat line, slope ≈ 0.

Here's what the 19 white-noise videos actually do:

The bass band is the loudest (−4.7 dB on average). The "air" band — everything from 12 kHz up — is forty-eight decibels quieter. That's not a small roll-off; that's a steep wall. Forty-eight dB is roughly the difference between a normal speaking voice and a quiet whisper.

Nothing called "white noise" should look like that.

Why are they so dark?

A few theories, in descending order of how much they likely contribute:

1. The recording chain doesn't go above 12 kHz

Many of these videos appear to be recordings of physical noise sources — fans, white-noise machines, hair dryers, vacuum cleaners — rather than mathematically generated noise. The microphones, room reflections, and post-processing all roll off the high end. By the time the signal hits YouTube's compression, very little energy survives above 12 kHz. The "air" band is just gone.

2. Deliberate softening for sleep

Pure white noise is harsh, especially through phone speakers or earbuds at low volume. Producers may be applying low-pass filters or EQ tilts to make the result more "comfortable" without telling anyone — trading spectral accuracy for pleasantness. We respect the goal. We just think it ought to be labeled differently.

3. YouTube codec behaviour at low bitrates

YouTube compresses audio aggressively. AAC at 128 kbps is the typical default for sleep videos, and AAC's psychoacoustic model is least kind to wideband stochastic content (which is exactly what noise is). The codec discards "imperceptible" high-frequency detail, which on a constant noise signal means a measurable high-end roll-off compared to the source.

4. The labels were always loose

"White noise" in colloquial use just means "constant masking sound." The mathematical meaning isn't intuitive to most people, and YouTube creators aren't required to validate their content against the textbook. The label is doing brand work, not engineering work.

Pink and brown: also off, but in different ways

Pink and brown follow the same pattern: noticeably steeper than spec. The pink videos average −7.56 dB/oct — closer to brown than to pink. The brown videos average −12.05 dB/oct, twice as steep as proper brown.

The headline takeaway: the YouTube "noise color" landscape is essentially shifted one full color darker than its labels suggest. What's labeled white is closer to pink-or-brown. What's labeled pink is closer to brown. What's labeled brown is darker than brown.

If you find the YouTube versions soothing, you've been listening to noise that's fundamentally different from the textbook spec. The spec isn't sacred — soothing is a real success criterion — but if you came in believing you were listening to white noise, you weren't.

The loop problem

The other thing we measured was repetition. Recording 8 hours of unique audio is expensive, so most "long" noise videos are short loops looped to fill the runtime. Our self-similarity analysis flagged a video as looped if it scored above 0.5 on a normalised cross-correlation between segments at different time offsets — meaning the audio repeats with detectable seams.

• White noise: 6 of 19 videos detectably looped, with loop periods averaging 29 seconds.
• Pink: 4 of 22 looped.
• Brown: 4 of 21 looped.

The white-noise loop rate (32%) is the highest, and the loop period is short — a 29-second cycle. Once you become sensitised to a 29-second loop, the masking effect drops sharply. Your brain stops perceiving a wall of static and starts perceiving a familiar audio file.

This is the strongest single argument for procedural generation over recorded loops, especially for long-form sleep listening. A procedurally synthesised noise stream has no loop point. There's nothing for your brain to lock onto.

The mono problem

Stereo white noise should have its left and right channels nearly uncorrelated — a stereo correlation around 0. We measured the left/right correlation for all 19 white-noise videos and found:

• Mean stereo correlation: 0.63 — well above what genuine stereo white should produce.
• 6 of 19 videos had correlations above 0.95, meaning they're effectively mono signals delivered in a stereo container.

That's not a small detail. Mono noise loses the spatial fullness that makes stereo noise feel like a wall around you. It collapses into a point in front of your face. For headphone listening especially, mono delivered as fake-stereo is a meaningful audio compromise. Genuinely stereo white noise (two independent uncorrelated streams) sounds noticeably wider and more enveloping.

What this means for sleepers

If you've been falling asleep to YouTube "white noise" successfully for years, this post isn't an accusation that you've been wrong. Whatever spectral curve those videos have, your nervous system has accepted it as a sleep cue. That conditioning is real and valuable.

It's worth knowing what you're actually listening to, though. Three things change when you understand the spectrum:

1. Switching to truly white noise will sound different. Mathematically pure white feels brighter, hissier, and more "active." It can take a week or two to adjust if you've been on YouTube videos for a long time. Many people prefer pink or brown after they understand the difference.
2. The loop seam may be the reason your sleep noise stopped working. If a video that used to mask your environment slowly stopped doing so, it's possible your brain has learned the loop. Procedural noise doesn't have this failure mode.
3. The mono-as-stereo issue degrades headphone listening. If you sleep with earbuds, an actually-stereo source (or a procedural source generated as two independent channels) gives you a fuller-sounding noise floor.

How to listen to actually-white noise

If this post made you want to hear the real thing, NoiseMoon's white preset is mathematically pure: a constant-power, flat-spectrum stream that's recomputed every sample on your device. The slope is 0 dB/oct by construction. There is no loop point. The two channels are independent — genuinely stereo, with correlation near 0 by design.

If pure white feels too bright (which is a fair reaction), use the warm slider to dial it darker. The slider is labeled warm ↔ bright; sliding it warmer rolls off the highs in a controlled, declared way — you choose how much, and the page tells you what you got. The defaults sit around −3 dB/oct (proper pink), which most people find more comfortable for overnight listening.

If you came here because the YouTube videos stopped working for you, try a procedural source. Your brain hasn't memorised it yet because nothing exists to memorise.

Methodology

Here's the recipe, briefly. Each video was streamed in audio-only mode and a 60-second segment from the middle of the runtime was extracted as 16-bit, 48 kHz stereo WAV. Mono down-mix was used for the spectral fit; the original stereo file was retained for left/right correlation analysis.

For the slope, we computed a long-window FFT (8192 samples, Hann windowed, 50% overlap), averaged the magnitudes into third-octave bands from 20 Hz to 20 kHz, and ran a least-squares linear fit to the band power in dB versus log-base-2 frequency over the 80 Hz to 8 kHz range. The slope of that fit is the reported dB/oct value. The fit's R² is reported alongside; for the 19 white-noise videos, mean R² was 0.91, meaning a single linear slope describes most of the spectrum well.

For the loop check, we computed normalised cross-correlation between non-overlapping 5-second segments of the recording, scanning offsets from 0.5 to 60 seconds. Any peak above 0.5 was flagged as a loop, and the offset at the peak was reported as the loop period.

For the stereo correlation, we computed the Pearson correlation between the left and right channels over the full segment. Genuine independent stereo noise gives values near 0. Mono signals delivered as duplicated stereo give values near 1.

The full data set (per-video tables, per-color summaries, raw spectrum CSVs) lives in the project repository at github.com/noisemoon/analysis. Replications welcome — if your numbers come out different, we want to know.

Closing thoughts

None of this is meant to dunk on YouTube creators making sleep noise content. The category is a real industry serving a real need, and there are talented producers in it who do good work. Many of the videos we measured almost certainly help people sleep. Whether the spectrum is technically correct is a different question from whether the audio is useful.

What we wanted to know is: when something is labeled "white noise" and serves millions of viewers, does the label match the spec? In the data we collected, the answer is mostly no.

If you found this useful, consider following along as we publish more analysis. Or just press play on Mira. Procedural white noise, generated on your device, never repeats, never streams. That's the alternative.