A video can reveal a location even when its picture offers few clues and its GPS metadata is absent. In a demonstration published November 2, 2026, the creator Colsto used faint aircraft sounds in a recording to narrow down where a security camera had been placed. The result is a striking example of software combining signal analysis and public data. It is also a specific demonstration, not proof that any uploaded video can be located automatically.
How can an aircraft sound point to a video location?
A moving aircraft changes the pitch of the sound reaching a stationary microphone. That Doppler pattern can appear as a curve in a spectrogram even when a listener does not notice the aircraft. Colsto compared patterns in the recording with historical ADS-B flight tracks, which record aircraft positions and times. Software modeled how candidate flights would sound from candidate ground locations, then ranked the matches.
The main recording contained two useful flyovers separated by about 15 minutes. Colsto knew the filming date and searched across the continental United States. The video shows the sound-based result being checked against the scene and a visible camera in map imagery. Colsto also reports that a separate one-minute segment with a known recording time ranked the same area first. Those are the creator's reported outcomes, based on a video demonstration; they are not an independent accuracy study.
Does this work on every video?
No. The recording needs a usable aircraft signal, historical flight coverage and enough information to limit the search in time or space. A single faint sound, an unknown date and an enormous search area make the task harder. The result also depends on the quality of the recording and the physical model used to compare it with the flight tracks. Colsto describes months of work and says the original code and tools are not being released, so the video does not offer a one-click way to locate arbitrary footage.
An independent project called DOPPLERGEIST now offers a source-available Python tool and local browser interface for a similar kind of analysis. Its documentation is unusually clear about the distinction between a ranked candidate and a confirmed location: a constrained real-audio example placed its top candidate 476 metres from an inferred reference, but the reference was not surveyed and the result was inconclusive. Its maintainers say independent blind accuracy on real recordings has not been established. DOPPLERGEIST is not Colsto's code and uses a noncommercial source-available license, rather than an open-source license.
What does this mean for people sharing software demos?
The practical lesson is that media can carry clues beyond the visible frame and file metadata. A maker posting a screen recording, game clip or app walkthrough may unintentionally include background sound from a home or workplace. That does not mean an aircraft analysis is likely to be used against every creator. Colsto himself says there are usually easier ways to find someone. It does mean that removing a location tag alone cannot guarantee that a recording gives away nothing about where it was made.
Before publishing a demo, consider whether the original audio is needed, whether it includes private conversation or surroundings, and whether the clip exposes a place you intended to keep private. A clean product capture can still show real software and its maker without relying on ambient sound. This is relevant to VIBPS as a social network for software: creators can share what they made while choosing what personal context to make public. VIBPS has not tested Colsto's method or DOPPLERGEIST for this article.