Cheap drones broke the economics of air defence. A few thousand dollars of hobby airframe can close an airport or destroy a tank, and the conventional sensors built to stop it are expensive, easy to blind, or both. Sound is one of the few signals a drone cannot switch off, because the propellers have to move air to stay up.
This project builds a garden-scale detector on that idea: a single cheap microphone, a pretrained audio model, and a persistence rule that decides when to alarm. The pipeline itself is deliberately unremarkable. The part that took the work is the evaluation, tested on drones the model never heard and recorded through other people's microphones, with the bars fixed before scoring anything.
I am writing it up as a four-part series, from the problem through to putting real hardware in my garden. The method, the full evaluation including the misses, and the code behind every figure go public as the series runs.
