For most of the smartphone era people have asked me “when will I be able to just hold my phone up and have it tell me what birds I’m hearing, you know, like Shazam does with music?”. “Ha!” I would chuckle derisively, “maybe never. While we have that capability for music, it’s much too computationally difficult to do that with birds”. Well, it’s finally happened – machine learning got way better, and some really smart people figured it out. The best of the suite of automatic bird song identifiers is now available to all, for free, via Cornell’s already indispensable Merlin birding app.
If you don’t know what Shazam is, then you’re about 23 years behind the times. The developers of this song-identification app, for music, created it in 1999. This is back when there was no such thing as an app. You literally had to call a number, from your land-line, or your Nokia candy-bar phone if you were a real techie, then hold your phone up to the song you wanted identified. I learned about it many years later when it was a well-known smartphone app, capable of identifying a song playing in the background of a noisy restaurant. Because of Shazam, people have been expecting the same app for bird songs, but years passed with no one getting any closer.
I won’t bore you with the technical details of how machine learning works, or what Deep Neural Networks are, mainly because I don’t understand them. But the capabilities of various forms of artificial intelligence have increased exponentially in a few short recent years, making possible things like this bird song identification and the photo identification algorithms behind the various plant identification apps, Google Lens, and my current obsession, iNaturalist.
In Merlin, which is a free, excellent all-around digital field guide for birds all around the world, you simply add the Sound ID module, open it, and tap the microphone. Instantly, it is telling you what birds are singing around you. If there’s one superpower I have in this world, it’s identifying birds by ear. And Merlin, while not perfect, is mind-boggling good. It currently can identify almost 700 North American and European species, with more being added soon.
I’ve watched it instantly pick up difficult birds like Bay-breasted Warbler and American Redstart, both tricky singers that I struggle with every spring. I saw it pick up a Yellow-bellied Sapsucker, uncommon in my neighborhood, calling on the next street over. The app helps you learn who is singing by continuing to light up the birds each time it detects them singing again. Tap on the species to hear other recordings and see photos, descriptions, and range maps.
Whenever I test the app, I feel like I’m in a way, way nerdier version of the old John Henry folk tale of man versus steam drill – I’m a person battling an imposing and mysterious new technology with nothing but my brain, ears, and eyes. I feel relieved when the app gets tripped up, as it did this morning – it was fooled by one of my local catbirds who likes to imitate the song of an Eastern Towhee. And I’m still an order of magnitude better with distant or quiet birds.
Just like printed guides and every other birding tool, Merlin Sound ID has been badly misused by a few novices who perhaps assumed it was infallible, or who didn’t have their phone GPS on, depriving the app of the context clues it needs to zero in on the answer. As a result, some wildly incorrect species have been reported recently here on the Cape, citing the app. So if you use it, don’t assume the app is always correct, because, just like human birders, it isn’t. Use it to help you learn how to identify the birds, rather than depend on it as a crutch. And make sure location services is on, so the app knows which birds are possible.
Meanwhile, I’ll continue my John Henry-style “man versus machine” battle with Merlin Sound ID. So far, I’m winning. I just hope I live long enough to hear the folk song they’ll inevitably write about me.
