Replies: 1 comment 2 replies
-
Those are likely false positives. There have been numerous reports in the BirdNET Analyzer Reddit community that the BirdNET model often misclassifies man-made sounds like police sirens as Eastern Screech Owl. Threshold values shown on your screenshot are so low that BirdNET-Go's dynamic threshold has been activated for that species, see https://github.com/tphakala/birdnet-go/wiki/BirdNET%E2%80%90Go-Guide#stage-2-confidence-threshold At the moment there is no way to control dynamic threshold per species, but what you could do is reduce the dynamic threshold expire time to, say, 1 hour in Settings > Main > BirdNET Settings. Also make sure that the trigger threshold is at least 0.9. In terms of what is different between BirdNET and Merlin: They have completely different AI models. BirdNET has made their model available, but Merlin has not. Also, it seems that BirdNET development has stalled as there have been no updates to the model since 2023. It is possible to train a custom classifier for BirdNET though. I have been training my own classifier to improve dog bark detection and reduce false positives caused by dog barks. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Long term fan and user of Birdnet-PI and now BirdNET-Go (Raspberry Pi 3B - default settings)!
Have been seeing the Eastern Screech owl show up consistently over the past week.
Audio files from these calls are usually less than 50% for the detection so I'm assuming they aren't nearby. Don't hear anything resembling a screech owl on the recordings.
I have tried on numerous evenings to see if I could personally hear or record them on Merlin. Had no luck whatsoever.
Just wondering if there's something different between Birdnet-Go and Merlin in terms of detection.
Thanks in advance,
Jim
Beta Was this translation helpful? Give feedback.
All reactions