BirdNET

BirdNET converts a short recording into a ranked set of possible species. The mobile workflow records sound, draws a spectrogram and applies a neural model before presenting confidence-linked suggestions. Browser demonstrations allow uploaded audio to be explored, while local and embedded versions support field stations and researchers working through much larger sound archives. Location and week-of-year models narrow what is plausible, and citizen-science submissions can contribute to broader bioacoustic research.

The score is evidence from a classifier, not an expert determination. Wind, insects, human noise, overlapping singers and distant calls can generate false positives or hide the target. A geographic prior may suppress an unusual but real species, and high confidence does not replace checking the recording against known vocalizations. The project explicitly warns against using its output alone for legal or conservation decisions. BirdNET is most productive as a triage instrument: it points to moments and names worth reviewing, while a person retains responsibility for the final identification.