Upload a field recording and separate overlapping birds, background noise, and ambient sound into individual playable layers. Then pick the one with your bird and download it.
- Free
- No account required
- Deleted automatically
- Works with phone recordings
- WAV, MP3, M4A, AAC, FLAC, OGG
Separate a recording
Up to 60 seconds, up to 50 MB, one recording at a time. Your recording is processed privately and deleted automatically. We do not use it to train models.
Loading the separator…
Hear it work
A real separation of a real Wings & Whistles field recording. Press play, then flip between Original and Isolated — the playhead does not move, so you hear the same instant both ways.
Wood Thrush and Blackpoll Warbler in a forest interior
A clean separationTwo birds singing in the same ten seconds, in a quiet forest interior at Rockefeller State Park. The Wood Thrush's flute-like song dominates; a Blackpoll Warbler's thin, needle-high trill sits above it.
What worked. The thrush and the warbler land on separate layers. Layer 1 is the Wood Thrush almost alone; Layer 2 is the high trill with the thrush's low notes pulled back.
What stayed imperfect. Layer 3 holds most of the forest ambience but still carries a shadow of the thrush. Nothing here is a clean studio recording, and it should not be treated as one.
2 of 4 layers cleared our confidence threshold. Excerpt from Wood Thrush and Blackpoll Warbler in a Forest Interior, Rockefeller State Park, recorded by Nick Paolino in the Rockefeller State Park. Separated in 9.7 seconds.
More examples, good and bad
We are showing you a hard case on purpose. Separation is genuinely useful and it is genuinely imperfect, and you should know which is which before you trust a result.
Wood Thrush and Blackpoll Warbler in a forest interior
A clean separationTwo birds singing in the same ten seconds, in a quiet forest interior at Rockefeller State Park. The Wood Thrush's flute-like song dominates; a Blackpoll Warbler's thin, needle-high trill sits above it.
What worked. The thrush and the warbler land on separate layers. Layer 1 is the Wood Thrush almost alone; Layer 2 is the high trill with the thrush's low notes pulled back.
What stayed imperfect. Layer 3 holds most of the forest ambience but still carries a shadow of the thrush. Nothing here is a clean studio recording, and it should not be treated as one.
2 of 4 layers cleared our confidence threshold. Excerpt from Wood Thrush and Blackpoll Warbler in a Forest Interior, Rockefeller State Park, recorded by Nick Paolino in the Rockefeller State Park. Separated in 9.7 seconds.
Red-eyed Vireo over a cricket chorus
A clean separationA Red-eyed Vireo delivering its unhurried, rambling phrases over a continuous cricket chorus. This is the classic problem case: the bird is not quiet, but the insects never stop.
What worked. The vireo's phrases and the cricket bed land on different layers. On Layer 1 the phrasing is far easier to follow.
What stayed imperfect. No species label is shown here. The classifier returned an unresolved internal code rather than a presentable name, so the abstention gate suppressed it, even though the separation itself worked. Separation succeeding and identification succeeding are two different things.
No layer here cleared our confidence threshold, so no species is named. Excerpt from Red-eyed Vireo Over a Cricket Chorus, recorded by Nick Paolino. Separated in 9.7 seconds.
A four-bird shrubby edge chorus (a difficult one)
A difficult oneA busy shrubby edge with a Blue-winged Warbler, a Common Yellowthroat, a Red-bellied Woodpecker, and a Baltimore Oriole all active. We include this deliberately: it is the kind of recording separation handles least well.
What worked. The Blue-winged Warbler's buzzy bee-buzz comes out clearly on Layer 4, well away from the rest of the chorus.
What stayed imperfect. The same warbler also bleeds into Layer 2, where it competes with the Common Yellowthroat and neither can be labelled confidently. The woodpecker's churr on Layer 3 scored below our label threshold and is deliberately left unlabelled. A bird appearing across more than one layer is normal, not a bug.
1 of 4 layers cleared our confidence threshold. Excerpt from Shrubby Edge Chorus with Oriole, Yellowthroat, and Blue-winged Warbler, recorded by Nick Paolino in the Rockefeller State Park. Separated in 9.5 seconds.
What separation can and cannot do
The credible promise is not perfect restoration. It is this: we will try to make the bird easier to hear.
What it does well
- Pulls a bird away from steady background sound: traffic rumble, insect chorus, distant machinery.
- Splits two birds that occupy different frequency ranges, like a low flute-like thrush under a high thin warbler.
- Makes a faint bird audible by moving the loud stuff somewhere else.
- Gives you every layer, so you can judge for yourself instead of trusting one automatic answer.
What it cannot do
- It does not invent audio. If the bird was not captured, nothing will bring it back.
- It does not reliably separate two birds singing in the same frequency range at the same instant.
- It does not always keep a bird on one layer. The same song can be split across several.
- It does not produce a clean scientific reference recording. Expect artifacts, especially on faint or heavily overlapped birds.
- It usually cannot tell you the species. See below.
Why we usually will not name the bird
We run a species classifier on every separated layer. We show you the result only when it clears a strict bar: at least 75% confidence, and a wide margin over its own runner-up.
That bar exists because classifiers are unreliable on separated audio. Testing this tool against our own recordings, the classifier confidently offered a warbler species that is not on the recording at all. A wrong bird, delivered with a straight face, is worse than no bird. So the default is silence.
When you see no species name, the separation still worked. Hearing the bird clearly and knowing its name are two different problems, and this tool only promises the first one.
What people use it for
The bird you could not hear back
You heard something in the woods, your phone barely caught it, and playback is all wind and footsteps. Separation pulls the bird off the noise.
Two birds at once
A dawn chorus where three species overlap. Split them onto separate layers and take them one at a time.
A cleaner clip to ask about
Before you post to a birding group or forum, isolate the bird so people are actually listening to the bird and not to a passing truck.
Teaching a sound
Educators and naturalists use it to demonstrate one vocalization hidden inside a complicated field recording.
Frequently asked questions
Is the Birdsong Isolator free?
Yes. It is free, there is no account, no signup, no payment, and no watermark on the audio you download.
Are my recordings private?
Yes. Your upload is processed privately and is never published. We strip the filename before it reaches our servers, we never give your audio a public URL, and the recording and its separated layers are deleted automatically within an hour. We do not use your recording to train models.
Can it separate two birds singing at the same time?
Often, yes, and that is what it is best at. Two birds in different frequency ranges usually land on different layers. Two birds singing in the same range at the same moment are much harder, and the same bird can end up spread across more than one layer.
Can it remove traffic, wind, or insect noise?
It separates them onto their own layers rather than deleting them, which is usually what you want. Steady sounds like traffic rumble and a cricket chorus separate well. Wind buffeting directly on the microphone is the hardest case, because it smears across the same frequencies as the bird.
Can it identify the bird?
Sometimes, and only when it is confident. We run a classifier on each separated layer, but classifiers are unreliable on separated audio: in our own testing one returned a species that was not in the recording at all, with meaningful-looking confidence. So we only show a name when it clears a strict confidence and margin threshold. Most of the time we show no name, and that is deliberate.
Does it change or invent the bird sound?
No. The tool reveals, it does not recreate. There is no synthetic bird replacement, no hallucinated reconstruction, and no artificial ambience. What you hear on a layer was already in your recording.
Why does the same bird appear in more than one layer?
This is normal. The separation model splits a recording into a fixed number of layers, and a single bird can be divided across several of them, especially if it changes pitch or overlaps another sound. Listen to every layer and pick the best one.
Why does the result sound distorted?
Separation introduces artifacts, particularly when the target bird is faint or heavily overlapped. The result is not an untouched scientific representation of your source recording, and it should not be treated as one.
Does it work on phone recordings?
Yes. Phone recordings are the main thing it is built for. Upload a WAV, MP3, M4A, AAC, FLAC, or OGG file up to 60 seconds and 50 MB.
Can I download the result?
Yes. You can play every layer, compare each one against the original, choose the one with your bird, and download it. Downloads are free and unwatermarked.
Can I upload copyrighted recordings?
Only upload recordings you have the right to process. You confirm this before a recording is separated.
Can I use the isolated clip for identification or research?
You can use it to hear a bird more clearly, and to produce a cleaner clip to send to a birding group or forum. Treat it with care as evidence: separation can both remove real sound and introduce artifacts, so the original recording remains the authoritative source.
What happens to your recording
- No account, no signup, no email.
- Your filename never leaves your device. We upload audio bytes only, and we never log filenames or audio content.
- Your upload and its separated layers are never given a public URL.
- They are deleted automatically within an hour.
- We do not use your recording to train models.
- You confirm you have permission to process a recording before it is separated.
Read the full privacy policy.
Want to understand what you are hearing?
Wings & Whistles teaches bird sounds through listening, comparison, and real recordings — the same field recordings the examples on this page came from. Start with how to identify bird sounds by ear, browse our Sound Moments, or get the app.