Acoustic UAS Recognition and Alerting
Start with sound.
Show the evidence.
A first trained baseline for recognizing drone-like acoustic activity in recorded audio. Listen to a sample, run the model, and inspect what it actually predicts.
01 / Choose a recording
Checking model availability…
02 / Inspect the prediction
PASSIVE AUDIO · EXPERIMENTAL BASELINE
The score is a model output, not a verified probability that a drone is present. This prototype does not establish aircraft identity, range, or field performance.
Initial held-out evaluation: 2 of 4 drone recordings detected; 0 of 24 background recordings flagged. One positive recording campaign; dataset-source effects remain unresolved. Read the full evaluation.
A real baseline
Predictions come from a fitted classifier. Training and evaluation use separate original recordings, with the data sources and limitations recorded alongside the model.
Public recordings, attributed
Initial sources: Salford’s DroneNoise database and selected ESC-10 recordings. Different recording environments can influence the scores.
The next evidence step
Independent recordings from new microphones and environments, followed by documented physical-sensor tests. Active sensing requires separate hardware and experiments.