I discovered a sound that leaks your exact location (and it's everywhere)

Main Ideas

• Background aircraft sounds can reveal location, even when they are inaudible. An aircraft’s Doppler shift can leave a recognizable trace in a recording’s spectrogram. By comparing that trace with flight data, the creator locates a security-camera recording despite the lack of useful visual clues.
• Doppler curves encode more than an aircraft’s presence. Their shape can indicate the aircraft’s speed and closest distance to the listener. Compared with the aircraft’s actual flight path, they can also show how it moved relative to the recording location.
• Silence is useful evidence. A location where two matching aircraft events occur about 920 seconds apart, with no other aircraft audible in between, is more plausible than a location near an airport where many other flights would be heard.
• The demonstration identifies the same location in two ways. A search using two events narrows down the recording’s location. A separate search using one minute of audio and a known recording time independently ranks the same location first. The creator then visually verifies the spot using Google Maps Street View.
• The technique is powerful, but not presented as an everyday certainty. The creator says the method could locate footage with no visual clues, while stressing that it is unlikely someone would use it to find a person and that easier ways of identifying a location often exist.

Methodology Shown

1. Find and measure the Doppler events

   • Inspect the audio in a spectrogram, with time on the horizontal axis and frequency on the vertical axis.
   • Identify the aircraft-related Doppler curves, including two events in the main recording.
   • Measure approximate frequency patterns and estimate ranges for each aircraft’s speed and closest slant distance, allowing for uncertainty.

2. Build a candidate flight dataset

   • Obtain ADS-B data containing aircraft positions, speeds, altitudes, and timestamps for flights over the continental United States on September 8.
   • Use aircraft type-code information to exclude known jet types, leaving potential propeller-aircraft flights.

3. Estimate when each location would hear each flight

   • Lay out a grid of candidate observer locations at 200-meter spacing.
   • For each location and flight, find the aircraft’s closest point of approach (CPA).
   • Convert the CPA into the delayed CPA time: when the sound would reach the observer, accounting for sound-travel time.
   • Use the aircraft’s altitude and a digital elevation model (DEM) for ground height to calculate three-dimensional slant distance, rather than relying on horizontal distance alone.
   • Initially consider aircraft within a conservative 10-kilometer radius.

4. Filter candidates using the two events and the silence between them

   • Keep locations where one suitable aircraft would be heard for the first event and another suitable aircraft about 920 seconds later.
   • Allow a few seconds of tolerance because event timing and measurements are imprecise.
   • Further filter by the speed and closest-distance ranges estimated from each Doppler curve.
   • Exclude locations that would have heard another flight during the roughly 15-minute interval between the recorded events.
   • These early filters reduce the candidate set from about 73,000 points to roughly 15,000.

5. Compare recordings with curves modeled from actual flight paths

   • Model what each candidate location would hear from the recorded ADS-B trajectory, rather than trying to infer a complete flight path from the audio.
   • Calculate the aircraft’s radial velocity—the component of its movement toward or away from the observer—and use it to predict the Doppler-shifted frequency.
   • Account for sound-travel time: the sound heard at a given moment was emitted earlier. The model therefore uses the aircraft’s position and radial velocity at the time the sound was emitted, not simply when it reaches the listener.
   • Compare each modeled curve with the observed spectrogram by summing brightness along the curve.
   • Sweep the event placement and fundamental frequency to account for measurement error. The relative timing between the two modeled events remains fixed by the flight geometry, so a mismatch in their offset counts against a candidate.
   • Refine the search from a 200-meter grid to about 100-meter spacing for the computationally intensive final scoring.

6. Check the result independently

   • Rank candidate locations by score, grouping results by flight pair so the best point for each pair can be compared.
   • Repeat a separate search using one faint Doppler curve, a known recording time of 8:37 p.m. UTC, and a search covering the United States.
   • The one-event and two-event searches rank the same location first. The creator then compares the result with Street View and identifies the security camera that recorded the footage.

Privacy and Mitigation Points

• High-pass filtering is not a dependable fix. The relevant frequencies overlap with those important to human speech. The creator demonstrates that processing and exporting audio as an MP3 can make a Doppler curve seem to disappear, only for it to become detectable again when frequencies are boosted.
• Louder foreground sound may not mask the clue. A demonstration plays a Doppler curve from a speaker while the creator talks, suggesting that ordinary speech or other loud sounds do not necessarily remove the background information.
• Phone-call compression may not remove it either. A demonstration suggests that the curve can survive phone-audio processing.
• Cuts can disrupt the timing evidence. Editing audio into separate pieces can prevent a search from using the time offset between two events.
• Additional noise cancellation is suggested as a possible measure, but the video does not establish that it will always work.
• The creator presents the video as an illustration of a capability, not as a claim that people are routinely located this way.

Speakers and Sources Featured

• Main speaker and narrator: The video’s creator, who explains the analysis and conducts the demonstrations. A brief phone-audio segment appears to feature the narrator speaking through a phone connection; no distinct additional speaker is clearly identified in the subtitles.
• Data and tools: ADS-B flight data, an aircraft type-code lookup table, digital elevation model data, spectrogram analysis, Doppler-equation modeling, and Google Maps Street View.
• Audio and visual material: The example security-camera recording, aircraft sounds and Doppler curves, and background music.