Looks very nicely implemented. I’m excited to give it a spin – and curious how to most effectively leverage their success for audio IDs here on iNat.
https://www.nytimes.com/2021/07/26/opinion/cornell-birdsong-id-app.html
I’ve excerpted the heart of Margaret Renkl’s article:
Though Merlin doesn’t claim to be 100 percent accurate, it comes very close. Drawing on a database of notes and recordings contributed by tens of thousands of citizen scientists through the Lab’s eBird initiative, Merlin listens as you listen, in real time, and tells you what you’re hearing. The app can identify some 400 North American species so far and will keep expanding. It’s an immense achievement, a quantum leap forward, nothing less than “a Shazam for bird songs,” as an article in Fast Company put it.
Naturally, I had to try out the new technology. I am far from an expert birder, but I do know my avian neighbors, and I figured a good way to test Merlin’s accuracy was to try it with birds I can already recognize by ear.
The first test didn’t bode well. I was reading on the sofa when I heard a Carolina wren singing just above my head. It was hopping around in a hanging basket barely a foot beyond the glass and singing its head off. That wren was as close as any bird was ever going to get, but the app was stumped. “Merlin has no matches,” it reported. Merlin fared no better in the two other recordings I made indoors.
But outside, something magical happened. I set my phone down on the table on my back deck, opened the Merlin app, chose “Sound ID” and hit the microphone button. Immediately a spectrogram of sound waves began to scroll across the screen. Every time a bird sings, the sound registers as a kind of picture of the song. By comparing that picture with others in its database, the app arrives at an ID.
I watched as Merlin rolled out the names of bird after bird — tufted titmouse, European starling, Carolina chickadee, northern cardinal, American crow, white-breasted nuthatch, eastern towhee, house wren, American goldfinch, blue jay, eastern bluebird, American robin, Carolina wren, house finch. It didn’t miss a single one.
What amazed me was not merely the accuracy of the ID but also the way the app untangled the layers of song, correctly identifying the birds that were singing in my yard, as well the birds that were singing next door and the birds that were singing across the street. If the same bird sang a second time, the app highlighted the name it had already listed. Watching those highlights play across the growing list of birds was almost like watching fingers fly across a piano keyboard.
Then I started seeing the names of birds I’d never seen in this yard before, birds that for me have existed only as undifferentiated sounds in the trees: Kentucky warbler, blue-gray gnatcatcher, yellow-breasted chat. The new bird I’d been hearing but not seeing all summer long, the one whose song sounded to me like, “Here, here, do you know my name?” turned out to be a magnificent summer tanager. Merlin also picked up the song of a yellow-throated warbler, a bird the app identified as uncommon for this area. I knew two were here because one of their babies fell out of a nest onto my son’s car — it was safely reared by the wildlife experts at Walden’s Puddle and released back into the wild — but I had never heard them sing. At least, I didn’t know what I was hearing when I heard them sing.