Removes the clues in a video's sound and picture that can reveal where and when it was recorded: passing planes, trains and ships, mains hum, flickering lights and metadata. Runs entirely in your browser; nothing is uploaded.
DETECTED LEVELS
| WHAT A TRACKER FINDS | IN → OUT | # |
|---|
average level · dB
spectrogram · frequency over time · example
stage log
HOW IT WORKS
The leak
Every plane, train and ship within earshot bends the pitch of its engine as it passes: a Doppler curve. Their positions are public (ADS-B, transit feeds, AIS), so someone can predict that curve for every place and time and check which one your audio matches. Matching works on tones far too faint to hear. Mains hum adds a second clue: its frequency drifts in a pattern that power grids log, which timestamps the recording.
The picture has a clock too
Lights on mains power flicker at twice the grid frequency. A phone camera reads its rows top to bottom, so the flicker lands as faint horizontal bands, and their drift is the same grid timestamp the hum carries. Each frame is scanned for those bands. If they're found, they're divided out of every row (you can untick that); it re-encodes the picture in standard dynamic range, so HDR footage looks a little different. Faint bands that barely move (60 Hz lights filmed at 30 fps) may not be found.
What you keep
Unless you remove flicker, the video stream is copied as is. Voices and the overall sound stay. The audio becomes mono (multiple mics reveal direction), the file loses its metadata (location, device, timestamps) and gets a random name, since phone file names often contain the date and time.
What cloaking does
What it can't do
This is a proof of concept. At best it reduces location leakage from sound; it doesn't make you anonymous: the picture, the words spoken or your account can still give the place away. So far it's tested on synthetic recordings. Posting later helps too, because every hour multiplies the flights an attacker has to check. Threat model.