Python tool for aircraft Doppler audio geolocation using historical ADS-B flight data (Source-available)
Pinned Hermes comparison
User-supplied reference: mojomast/hermes-aircraft-doppler-geolocation,
commit 2c3ed20577bcce29058d73422dd4f85392d04c24, MIT license, Kyle Durepos.
It is another independent implementation of the public video's method. No donor
code was copied into our implementation in this comparison.
The inspected scripts supply band/window-selected Viterbi ridge tracing, sub-bin measurement, multistart straight-pass fitting, explicit rejection diagnostics and emission-time candidate scoring. The README distinguishes real tone detection from its synthetic location self-test. It does not supply automatic event discovery, historical flight/terrain acquisition or national search. Its synthetic test supplies a single known flight and includes the true observer on its grid; that result cannot establish blind off-grid real localization. Its scorer's partial-overlap support is less strict than DOPPLERGEIST's full-event support.
Our local comparison command was
/home/cinna/aircraft-audio-geolocation/.venv/bin/python -m pytest -q tests
in /home/cinna/dopplergeist-reference-hermes: 6 passed in 19.64 s, including
the donor's offline synthetic extraction/scoring test. This is a reference-code
check, not DOPPLERGEIST automatic upload or national accuracy acceptance. The
integration board records another worker's retained full donor replay; do not
repeat it merely to duplicate green checks.
A useful remaining audio experiment is a seeded dynamic-programming continuation for weak ridges, reusing our existing audio.trace_ridge. Any such fallback must select its own seeds/bands from the recording, reject unsupported noise paths, and be evaluated at fixed settings. A successful manual-band donor result is not evidence that this automatic step already works. Turning-flight fit rejection must not become a compulsory discovery gate.