What the scrubber does
An aircraft engine hums at a steady pitch. As the plane passes, that pitch slides down: the Doppler effect. A recording keeps the slide as a faint curve, mostly between 50 and 300 Hz colsto, “I discovered a sound that leaks your exact location” (YouTube, 2 Nov 2026).
The scrubber looks in that low band for tones that last at least a couple of seconds and either fall smoothly or stay flat. Engines, mains hum and fans behave like that. Voices do not: their pitch jumps between syllables.
Each tone it finds is lowered to the level of the noise around it, frame by frame. Everything else in the recording is left as it was, so speech still sounds the same.
Mains hum is caught too. Its 50 or 60 Hz frequency drifts with the power grid, and forensic analysts use that drift to date recordings. The scrubber lowers the hum lines inside the band you choose; higher harmonics stay.
Why add decoys?
Decoys are fake flyovers: a tone with two harmonics, gliding down like a real pass. Each gets a random pitch, speed, distance and time, so no two files carry the same set.
On their own, decoys are weak. The real curves would still be in the recording, and an analyst can match them to flight records and ignore the fakes. That is why the scrubber removes the real tones first. Decoys then make a blank spectrogram look ordinary and waste the time of anyone matching curves.
What happens to a video
The picture is copied unchanged, without re-encoding. The cleaned soundtrack is encoded as AAC at 192 kbit/s, or Opus for WebM files.
The new file also drops the original’s metadata: creation time, device details and any location tag. To rebuild the video, the page loads ffmpeg.wasm (about 32 MB) from jsDelivr once. It checks each file’s SHA-256 fingerprint before running it.
How we tested it
We built a 60-second test recording at 48 kHz. It held speech-like sound, a quiet propeller pass (110 Hz, 70 m/s, 700 m away), 60 Hz hum and background noise.
On the Normal setting, the propeller curve went from 8.7 dB above its surroundings to 4.0 dB below them. The hum line was removed too, and a second check found no long tones left.
The speech-like sound lost about 0.3% of its energy. On speech alone, Normal and Gentle found nothing to remove.
We also rebuilt an MP4 and a WebM test video. The picture stream came out bit-for-bit identical, and the location, title and creation-time tags were gone.
Held musical notes in the low band are tones too, and Normal thins them. Use Gentle for music. We have not tested the tool against a real investigator, and no filter can promise that nothing is left.
What it cannot do
- It does not touch other sound clues: voices, accents, sirens, bells, birdsong, trains or announcements.
- It does nothing about what the picture shows, or about where and when you post.
- Very loud or very close aircraft can leave traces. Check the after picture, and try Strong.
- Trimming still helps most. Cutting out sections breaks the time gaps between flyovers that the method relies on colsto, “I discovered a sound that leaks your exact location” (YouTube, 2 Nov 2026).
For the full method and the other defences, read how plane sounds locate a recording. For photos, the EXIF viewer removes GPS tags.
Frequently asked questions
Can a sound in my video really reveal where I was?
Sometimes. In November 2026 a researcher matched faint aircraft Doppler curves in a clip to public flight data. He placed the camera within about 100 metres. It needs work and the right recording, but it is possible.
Does the Doppler Scrubber upload my file?
No. The analysis runs in your browser. For videos the page downloads the ffmpeg.wasm library from jsDelivr, but your file is never sent anywhere.
Will it change how my voice sounds?
Normally not. It only lowers long, smoothly falling or steady tones below 400 Hz by default. Speech pitch jumps between syllables, so it is left alone. Held musical notes can be thinned; use Gentle for music.
Why not just use a high-pass filter?
Aircraft tones share frequencies with the human voice, so cutting the whole band cuts speech. The researcher also found that a filtered curve came back when the quiet band was boosted. The scrubber lowers only the tones, down to the background level.
Is adding fake Doppler curves enough on its own?
No. Real curves would still be there, and they can be matched to flight records while the fakes match nothing. Decoys help only after the real tones are removed.