This dataset provides real-world, open-field acoustic recordings of a DJI Air 3 drone captured with a custom ground-based 24-microphone array, synchronized with GPS flight telemetry. Recordings span multiple sessions across different days and two distinct outdoor locations, including drone-present and ambient "no-drone" segments used to calibrate noise-robust detection models. The dataset supports research in sound source localization (SSL) and Sound Event Localization and Detection (SELD), and was used to train and validate a U-Net-based model that reformulates DoA estimation as spherical semantic segmentation over delay-and-sum (DAS) beamformed acoustic energy maps.

Dataset contents: Multichannel WAV audio recordings (24 channels, 48 kHz, four synchronized Zoom F6 recorders) of a DJI Air 3 drone in open-field flight, plus CSV flight-log files (GPS position, altitude, speed, heading at 100 ms resolution). Includes per-session alignment parameters (JSON) and reference-flight recordings used to calibrate the GPS-to-array coordinate frame, plus ambient "no-drone" background audio. Two of the four sessions also include a 360-degree reference video (Insta360 X4). Companion Python scripts are included so others can regenerate the labelled dataset used to train the model.