Premium apps Part of Soundscape Atlas
Noise Generator (white, pink, brown)
Make white, pink, brown, blue or violet noise as a WAV file and check its spectrum.
Try an example: Pink noise, 5 s stereo, 44.1 kHz · Brown noise, 4 s mono loop at 48 kHz · White noise test signal, 2 s mono, 32-bit float
How it works
Generates coloured noise with the correct spectral slope (white 0, pink −3, brown −6, blue +3, violet +6 dB per octave) as a downloadable WAV, mono or decorrelated stereo, 8–96 kHz, with level, fades, seamless looping and a seed for repeatable files. The measured spectrum, slope and levels are reported so you can verify the colour.
- White noise: Gaussian random samples from a seeded sfc32 generator (Box–Muller), so the power spectrum is flat.
- Pink noise: white noise through a cascade of first-order pole/zero filters designed for the chosen sample rate (a pole every two octaves from 5 Hz, a zero half-way between), which averages −10 dB per decade = −3 dB per octave within about ±0.3 dB.
- Brown noise: a leaky integrator (corner 10 Hz, so it never drifts) gives −6 dB per octave; blue noise inverts the pink filter (+3 dB); violet noise is the first difference of white noise (+6 dB per octave below about a quarter of the sample rate).
- The level is set as RMS over all channels; if the highest peak would exceed −0.1 dBFS the gain is lowered and reported. Fades use an equal-power (sine) curve. Seamless loops render extra noise and crossfade it onto the start.
- Check: the result is analysed with a Hann-windowed 4096-point Welch spectrum; the slope is a least-squares fit over 1/3-octave bands from 100 Hz to 10 kHz (or a quarter of the sample rate).
Start playback quietly: noise at full level can damage speakers and hearing. Noise is not a medical sleep or tinnitus treatment.
Sources
- Colors of noise (spectral slopes of white, pink, brown, blue and violet noise) (Wikipedia)
- P. D. Welch (1967): The use of fast Fourier transform for the estimation of power spectra, IEEE Trans. Audio Electroacoustics 15(2) (IEEE)
- Multimedia Programming Interface and Data Specifications 1.0 (RIFF WAVE), IBM & Microsoft 1991 (McGill University (archived specification))
Part of Soundscape Atlas
-
Soundscape Atlas Sound map & ambience studio
Find real field recordings on a world map and mix them into calm, editable soundscapes.
Related tools
- Ambience Generator (rain, wind, stream, fire, hum, drone) – Synthesize rain, wind, water, fire, room hum or a drone as a seamless-loop WAV.
- Audio Spectrum Analyzer – Upload audio and see its spectrum, peak frequencies, centroid, RMS/peak and waveform.
- Universal Audio Converter – Convert MP3, WAV, FLAC, Ogg, Opus, M4A and AIFF with bitrate and sample-rate control.
How to use the Noise Generator (white, pink, brown)
Make white, pink, brown, blue or violet noise as a WAV file, set level, fades and looping, and check the colour in the measured spectrum.
What it is for
Pink noise is the standard test signal for speakers and room EQ because every octave carries the same energy. Brown noise is the deep rumble many people use to mask a noisy office or to sleep. White noise is the classic signal for testing filters and audio chains. This tool generates all of them, plus blue and violet noise, as a WAV file you can download and loop.
Each colour has a defined spectral slope: white 0, pink −3, brown −6, blue +3 and violet +6 dB per octave. The tool analyses the file it made and reports the measured slope, levels and octave bands, so you can check the colour yourself.
It is not a medical sleep or tinnitus treatment. For rain, wind and other textured sounds, use the Ambience Generator.
Worked example
For a background sound on a loop, we asked for brown noise, 30 s, 48 kHz, stereo, level −24 dBFS, seamless loop on and seed nap.
The summary: "Brown / red (−6 dB/octave) noise, 30 s stereo at 48000 Hz: measured slope −5.95 dB/octave (nominal −6), RMS −24 dBFS, peak −11.6 dBFS." The octave-band table falls steadily from 0 dB at 32 Hz to −16.6 dB at 1 kHz and −26.9 dB at 16 kHz, which is the brown slope. The loop check reported a largest jump of 0.002 at the seam, against 0.017 inside the file, so the restart is inaudible. The file is 5.76 MB (PCM 16-bit).
A second run shows the clip guard. White noise, 8 kHz mono, requested at −6 dBFS RMS, came back with a warning: "The requested level of −6 dBFS would have clipped, so the noise was made quieter (RMS −12.9 dBFS)." Gaussian noise has peaks about 13 dB above its average, so a loud RMS level cannot fit below full scale.
How to read the result
- Measured slope vs. nominal slope is the colour check. It is fitted over 1/3-octave bands from 100 Hz to 10 kHz (or a quarter of the sample rate). A value within about ±0.3 dB of the nominal slope is normal.
- RMS level is the average loudness of the file, including any fades. Peak level is the largest sample. The crest factor is the difference between them (about 12–14 dB for noise).
- Octave-band table: equal values mean pink noise. Rising values mean white, blue or violet noise, and falling values mean brown.
- Loop seam appears only for loops. A seam jump no bigger than the jumps inside the file means a click-free loop.
Tips and common mistakes
- Start playback quietly. Noise at −18 dBFS RMS through headphones turned up high is loud.
- Use pink noise for speaker and room measurements, brown for masking and sleep, and white for testing electronics.
- Keep the seed to recreate exactly the same file later. Change it if you need several different noise files of the same colour.
- For long listening, make a 30–60 s seamless loop and let your player repeat it, instead of a huge file.
- Choose 32-bit float only for editing software; 16-bit PCM plays everywhere.
- Very short files (under a second) give noisy slope estimates.
FAQ
Is pink noise made with the Voss-McCartney algorithm?
No. White noise goes through a chain of first-order filters designed for the chosen sample rate, with a pole every two octaves and a zero in between. That gives −3 dB per octave within a few tenths of a dB from 8 kHz up to 96 kHz sample rates.
Why are left and right different?
Stereo files use two independent random streams, so the noise sounds wide instead of sitting in the middle.
Will I get the same file again tomorrow?
Yes. The generator is fully deterministic: the same colour, settings and seed always produce the same bytes.
How long can the file be?
Up to 120 seconds, and up to 60 seconds for stereo at 96 kHz. The download stays available for about 2 hours.
Guide last updated:
For scripts and AI agents
- JSON API:
POST https://soundscapenoisegenerator.usefulweb.app/api/v1/calculate(same calculation as this page) - Interface description: OpenAPI 3.1 · AI guide: llms.txt
- MCP server (Streamable HTTP):
https://soundscapenoisegenerator.usefulweb.app/mcp· tool namesoundscape_noise_generator
Example request
curl -s -X POST https://soundscapenoisegenerator.usefulweb.app/api/v1/calculate \
-H 'content-type: application/json' \
-d '{"type":"pink","duration":5,"sampleRate":"44100","channels":"stereo","level":-18,"seed":"1"}'