Attracting Birds

Bird Song and Sound ID Apps

Most of the birds in your yard are heard rather than seen. Here is the vocabulary, a learning method that works, ten mnemonics for feeder regulars, and an honest account of what a sound ID app can and cannot do.

Stand in your yard for five minutes with your eyes shut and you will notice something uncomfortable: there are more birds than you thought, and you cannot name most of them. Sound is where the majority of birds actually are. They are inside a hedge, behind a leafed-out maple, two gardens over, or flying past 200 feet (60 m) up, and none of those birds appear on a feeder.

Learning bird sound is the single biggest step-change available to a backyard birder, larger than better binoculars and much larger than a better feeder. It is also the part people put off, because it feels like memorizing a language with no writing system. It is not that hard if you go five birds at a time and stop trying to learn everything at once.

This page covers the vocabulary you need, the dawn chorus, the mimics that break every rule, a learning method with mnemonics for ten feeder regulars, and then a straight account of how Merlin's Sound ID works, what it is genuinely good at, and where it fails. We are not going to quote an accuracy figure for any app, because no honest single number exists.

Why sound doubles your bird list#

In summer, foliage hides almost everything. A yard that shows you eight species at the feeder will often produce fifteen or sixteen to someone listening, and the extra ones are the birds that never come to seed at all: vireos, warblers passing through, a wren deep in the brush pile, a flicker calling from a dead limb.

Sound also tells you things a sighting cannot. A sudden universal silence at the feeder means a hawk has arrived, usually before you see it. The volume and pitch of a chickadee's alarm scales with how dangerous it thinks the threat is. And arrival dates, which is what makes bird sound valuable for citizen science, are usually detected by ear weeks before you get a clear look at anything.

The vocabulary#

Four terms cover most of what you need.

• Song. Long, structured, often repeated, usually given from a conspicuous perch, mostly in the breeding season, and in most North American songbirds mostly by males. Songs defend territory and attract mates.
• Call. Short, simple, given all year by both sexes. Contact calls keep a pair or a flock together, flight calls are given in the air, and begging calls come from young birds. Calls are often the more useful identification tool at a feeder, because feeder birds call constantly and sing comparatively little.
• Chip note. The short, sharp single call that sparrows, warblers and finches give continually while foraging. These are genuinely hard, and separating them is an advanced skill; a metallic chip from a northern cardinal is a good first one to learn because it is loud and distinctive.
• Alarm call. Given at a predator, and often graded, meaning the call changes with the level of threat. Other species listen to them, which is why one scolding chickadee can empty a yard.

The dawn chorus#

For a few weeks in late spring, singing peaks in the half hour before and after sunrise, then falls away through the morning. The likeliest explanations are that early light is too dim for efficient foraging, that still cool air carries sound further, and that a male advertising at first light is signaling that he survived the night. The evidence for any single explanation is not settled, and you do not need it to be.

What matters practically is that the dawn chorus is both the best and worst time to learn. Best, because every bird in the neighborhood is singing its full song. Worst, because they are all doing it at once, which is also the condition under which sound ID apps struggle most. Learn in the late morning, when one bird sings at a time.

The mimics#

A handful of species imitate others, and they will fool you and every app you own.

• Northern mockingbird. Runs through long strings of borrowed phrases, typically repeating each one three or more times before switching. It will also imitate car alarms, phone tones and squeaky gates. A single mockingbird can generate a dozen wrong species suggestions in a minute.
• European starling. Mimics with an odd wheezy, clicking, whistling delivery. Its impressions of killdeer, meadowlarks and mechanical noises are good enough to catch experienced birders out.
• Gray catbird. Similar material to a mockingbird but usually each phrase once, delivered in a rambling, disjointed way, punctuated by the cat-like mew that gives it its name.
• Brown thrasher. Repeats each phrase roughly twice. The old rule of thumb, thrasher twice, mockingbird three or more, catbird once, is imperfect but genuinely useful.
• Blue jay. Not a general mimic, but a very good imitator of the red-shouldered hawk's kee-yer scream, apparently used to clear a feeder. If you hear a hawk and everything scatters and then a jay drops onto the seed, you have been played.

How to learn: five at a time#

The method that works is small and boring. Pick five birds you already know are in your yard, learn them properly, and only then add five more. Ten well-learned sounds cover most of what you will hear on an ordinary morning, and thirty covers almost everything.

1. Choose the five commonest birds at your own feeder. Local and familiar beats comprehensive.
2. Listen to each one deliberately, several times, on a reference archive rather than on an app's live screen.
3. Say the mnemonic out loud. It feels silly and it works, because it gives the rhythm a handle.
4. Go outside and match sound to bird visually. Hearing a titmouse while watching the titmouse do it is worth ten recordings.
5. Test yourself with your eyes closed. If you can name all five without hesitating for a week, add five more.
6. Learn the calls as well as the songs. Feeder birds call far more than they sing, and the calls work all year.

Ten feeder-bird sounds worth learning first#

Two of these are worth extra attention. The chickadee fee-bee is the sound that separates black-capped from Carolina chickadees in the overlap zone, where photographs cannot. And the goldfinch flight call is how you learn that goldfinches are passing over your yard all day, which no amount of watching the feeder would tell you.

How Merlin's Sound ID works#

Merlin, from the Cornell Lab, is free and it is the tool most people mean when they say sound ID app. The mechanism is worth understanding, because it explains the failures.

Your recording is converted into a spectrogram, a picture of the sound with time along one axis, pitch on the other, and loudness shown as brightness. Merlin then runs an image classification model over that picture, which is the same broad approach used for photo identification. The model was trained on curated recordings, a minimum of around a hundred per species, trimmed and labeled by volunteers, and it uses eBird data to weight its suggestions toward what is actually likely at your location on that date. It currently covers a couple of thousand species, which includes most of the United States, Canada and Europe.

What it is genuinely good at is a clean recording of a single common bird singing at moderate range on a still day. In that situation it is a superb learning tool, because it shows you the spectrogram as the bird sings, names it, and lets you play a reference recording to compare. Watching the spectrogram is the part that teaches you the most, and it is the part people ignore.

Where it fails, and the rule that follows#

The failure modes are consistent enough to anticipate.

• Overlapping songs. A dawn chorus with six birds singing at once is exactly the hardest case, and it is when most people reach for the app.
• Wind, traffic, rain, air conditioning and lawnmowers. Broadband noise swamps the spectrogram, and quiet birds disappear into it.
• Distant or faint birds. A far-off call may produce nothing, or a confident and unrelated suggestion.
• Mimics. A mockingbird or a starling will generate a stream of plausible species names, none of which are present.
• Rarities. The model returns a name whether or not it should, and it has no way to express doubt in the interface. A confident wrong answer looks exactly like a confident right one.
• Non-birds. Squeaky doors, phone alerts, children, other people's playback and other people's phones all get classified as something.
• Similar species pairs that even experienced birders separate carefully by ear.

None of this is a reason to avoid the app. It is a reason to use it as a teacher rather than an oracle. Used well, it is the fastest way to learn sound that has ever existed, because it gives you an instant candidate to check against a bird you can go and look at. Used badly, it produces a yard list full of birds that were never there. The same reasoning applies to the automatic species labels on a camera feeder, and for the same underlying reason.

Macaulay Library and xeno-canto#

Behind the apps sit two large open archives, and both are worth using directly.

The Macaulay Library, run by the Cornell Lab and integrated with eBird, holds an enormous collection of bird recordings, photographs and video contributed by observers worldwide. Every eBird checklist with media adds to it. Its strength for learning is that recordings are tied to a location, a date and an identified checklist, so you can listen specifically to birds recorded near you rather than to a population from the other end of the continent, which matters for species with strong regional dialects.

Xeno-canto is a community-built archive of bird sounds from around the world, contributed and annotated by recordists and released under open licenses. Its coverage of variation is often excellent: alarm calls, flight calls, juvenile begging and regional variants that a field guide never mentions. Between the two you can nearly always find several examples of any sound you are trying to pin down.

Use them the same way you would use a field guide. Listen to several recordings of the same species rather than one, pay attention to the labels distinguishing song from call, and note where the recording was made. Then go outside and listen to the actual bird, which is the part that makes it stick. If you want to turn that skill into something useful, the projects in citizen science for feeder watchers accept sound records, and a recording is the same kind of evidence as a photograph in feeder photography.