FAQ¶
Frequently asked questions.
General¶
Q: Does BirdNET Live require an internet connection? A: No. All inference runs on-device using the ONNX model. The only network features are species image/description lookups from the taxonomy API, which are optional.
Q: How many species can it identify? A: The BirdNET+ V3.0 model identifies 5,250 bird species worldwide (the pruned intersection of the audio classifier and geo-model).
Q: What platforms are supported? A: Android (8.0+), iOS (15.0+), and Windows (experimental).
Accuracy¶
Q: Why is my confidence threshold showing low scores? A: Lower the confidence threshold in Settings to see more detections. Background noise, wind, and distance affect accuracy.
Q: What does the species filter do? A: The geo-model predicts which species are likely at your GPS location and time of year. Enable "Geo Exclude" to hide unlikely species, or "Geo Merge" to weight results by geographic probability.
Q: How accurate is the identification? A: Accuracy depends on recording quality, distance, background noise, and the species. High-confidence detections (>70%) are generally reliable. Always verify rare species visually.
Recording¶
Q: Where are recordings saved?
A: In the app's documents directory under recordings/<session-id>/. Full recordings are saved as WAV files.
Q: Can I analyze existing recordings? A: Yes. Open File Analysis from the home screen, pick an audio file, set location and parameters, and tap Analyze. Supported formats include WAV, FLAC, MP3, OGG, Opus, M4A, AAC, WMA, and AMR.
Point Count¶
Q: What is Point Count Mode? A: A timed survey mode for formal avian point-count observations. You set a fixed duration (3–20 minutes) and location, then the app runs continuously and auto-stops when the timer reaches zero.
Q: Can I pause a point count? A: No. Protocol compliance requires uninterrupted recording. You can end early via the stop button.
Q: Where do point count results go? A: They appear in the Session Library as "Point Count #1", "#2", etc. You can review, edit, and export them like any other session.
Performance¶
Q: Why is the app warm / using battery? A: ONNX model inference is compute-intensive. The screen also stays on during live sessions. This is normal for real-time neural network processing.
Q: The spectrogram looks frozen. A: Ensure microphone permission is granted and audio capture is active. Check that no other app is using the microphone.