Frequently asked questions
What BirdAmp does, what it costs, what happens to your recordings, and how much to trust an answer.
Questions
How accurate is BirdAmp?
It depends almost entirely on the recording. A close, clean clip of a single bird singing is usually identified correctly. A distant bird in wind, several birds at once, or a brief contact call is much harder. This is why every detection carries a High, Medium or Low confidence label rather than a bare species name — the label is telling you how much weight the result will bear.
Does it work without an internet connection?
Yes. The model ships with the app and runs on your device, so identification works with no signal at all. You only need a connection to install the app and to load species photographs.
Is my audio uploaded anywhere?
Not for identification. Recordings are analysed locally on your device and stay there unless you deliberately share or export them.
How many species does it know?
More than 6,000 species worldwide. Coverage is best for widely recorded species and thinner for those with few available recordings, which tends to mean rarities and birds from under-recorded regions.
Why does it ask for my location?
Location and time of year let the app weight the shortlist towards species that plausibly occur near you. It is optional, and identification works without it, but the results are noticeably better with it — a great many wrong answers are birds from the other side of the world.
Why was a High confidence result still wrong?
Confidence measures how strongly the audio matched what the model expects, not whether the bird is really there. Similar-sounding species, mimics such as Starlings, and unusual variations of a familiar song can all produce a confident mistake. Treat confidence as evidence to weigh alongside location, season and habitat, not as a verdict.
How long should a recording be?
Ten to thirty seconds, containing two or three clear repetitions. Longer is not better — a short clip with good song beats a long one with a single distant call.
What audio formats can I upload?
The common ones a phone or recorder produces: WAV, MP3, M4A, AAC, FLAC and OGG, plus the audio track from a video file. Uncompressed WAV or FLAC preserves the most detail.
Can I use recordings from a dedicated recorder?
Yes, and they usually work better than phone recordings. Anything that gets a better microphone closer to the bird will improve the result.
Can it separate several birds singing at once?
Up to a point. Neural Boost is designed to pull one bird out of a mixture, and a dawn chorus will often yield several species. But a busy recording is genuinely hard, and quiet birds behind loud ones are routinely missed.
Does it identify non-vocal sounds?
Not reliably. The model is trained on vocalisations, so things like woodpecker drumming or wing noise are outside what it does well.
Can it identify birds from photographs?
No. BirdAmp works from sound only.
What does Neural Boost actually do?
It re-synthesises a single detected bird from the mixture, so you can hear that bird more clearly on its own. It is both a listening aid and a way for the app to show its working — you can check whether what it isolated really is what it claims.
What should I do if an identification looks wrong?
Play back the boosted audio for that detection first: quite often the model has picked out a real bird you had not noticed, rather than misnaming the one you were listening to. If it is genuinely wrong, the feedback option in the app records the correction.
Does it cost anything?
The core identification is free to use. Some additional features are offered through an optional subscription.