SpectroBird: Event-Level Vocalization Detection and Curation for Avian Ecology
Passive acoustic monitoring often summarizes long recordings as fixed-window presence predictions, but many ecological questions require records of individual vocal events: how many occurred, when they occurred, whether they overlapped, how long they lasted, what frequency band they occupied, and whether each event was complete enough for structural measurement. We address this event-level…
SpectroBird is a spectrogram-based approach designed for detecting and curating individual avian vocalizations in noisy field recordings, providing a solution to the event-level measurement problem. The system applies a detection-localization model that identifies candidate vocalizations as time-frequency boxes, followed by an event-consolidation step that merges repeated detections from overlapping windows while treating overlapping vocalizations as separate events.
A completeness-curation model then classifies each event as complete or incomplete before allowing for downstream event-level analysis.
To demonstrate the effectiveness of SpectroBird, it was tested on breeding-season Hume's Leaf Warbler buzz vocalizations recorded at varying elevations in the Himalaya. The detection-localization model performed comparably to AST and YOLOv11-cls baselines while maintaining accurate event timing and frequency bounds. Event-level evaluation achieved an impressive AP of 0.999 for complete buzzes.
When applied to two field seasons, the resulting event records revealed that the duration of complete buzzes decreased with elevation, showed no simple elevation trend in occurrence, and the relative day-of-year occurrence patterns were strongly correlated between years. These findings highlight the potential of event-level detection and curation using SpectroBird to support ecological analyses that rely on individual vocalizations rather than fixed-window presence alone.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
