The New Merlin Bird ID Shazam for Birds ID will identify that spot for you
I was recently walking through a clearing of felled trees in a wooded Brooklyn park with my iPhone in hand. The birds were singing everywhere, but through the roar, I was recording a particular song: It was almost certainly the forgotten, metallic whistle of a Bicknell’s thrush. Although a light-looking, brown-spotted bird, this rare thrush is the first target of New York birdwatchers – but its identification poses a challenge. Unless you hold it in your hand, you can’t identify it reliably based solely on its appearance, and its edge differs only slightly from its doppelganger, the most common gray thrush.
I left the woods with only a muddy register of experience, a mixture of background noise and the chirping of other birds. But when I loaded the file into the new Sound ID feature of the Merlin Bird ID app, it correctly named every bird in the recording, including cardinals and warblers, and could discern between the faint sound of Bicknells and the gray thrushes that were tramindui. on registration.
A lot of apps try to identify birds by images and sounds, with varying levels of success — an app they asked me to review called each recording a Nordic mockbird, a bird that mimics other birds. . But birders and citizen scientists have long relied on the Merlin ID of the Cornell Ornithology Label to help identify photos of birds. When I discovered that they had extended their services to bird watching, I was quick to try and eager to learn more about what lies behind the machine-powered sound identification.
Experienced birds can quickly identify birds with their unique songs, but doing so can be difficult and requires time and experience. Such is the purpose of Merlin Bird ID — to help those who are still trying to understand things. “The interesting thing about Merlin is that he’s an unassuming partner who can say you hear a sparrow song for the 300th time, and he’ll tell you with pleasure like the first time,” said Drew Weber, Merlin Bird Project Coordinator ID.
I picked up the app for a dedicated test drive this past Saturday in Prospect Park in Brooklyn to make sure its success over the previous recording wasn’t a fortune. Although the city’s location and ecology make it a prime destination for bird watching during spring and fall, only a few songbirds remain in the parks during the summer, so the app would have the advantage of having mainly common birds to be identified.
I stopped at a tree at the noisy southwest entrance of the park, where a Baltimore oriole was singing from a pine tree. I turned on the sound ID feature, set a record, and held my phone over my head. The app showed me a spectrogram – a graph of the frequencies it was recording over time – and immediately suggested “American Robin;” in fact, a robin had begun to sing behind me. I tried again, and this time, a house sparrow started crying. The app showed me the photo of a house sparrow. I tried one last time, and as soon as the oriole was singing, a thud sounded from above; the application replied that it had again ignored the oriole in favor of correctly identifying something else. I guess this showed the agility with which the app could offer identification, but I was frustrated that it failed to identify the oriole – a common bird – in this easy environment.
As I walked through the woods of the park, I kept the app open and registered for any other birds I might find. He successfully identified a “pew-p” of a northern cardinalew-pew “song, even as the cardinal began to make a high-pitched note, the app suggested amusingly that he was listening to an osprey, a huge hawk eating fish. Strong notes of the wax’s” sight ” of cedar appeared crisply on the spectrogram, even if the sound is not identified, and instead an image of a warbling vireo appeared as it began to sing in the distance (a song I heard it describe) as “a drunken person who tries to make a point “).
Merlin’s Sound ID won me over, though; I heard barely a couple of distant notes, and immediately the app suggested Acadian flycatcher, a southeastern forest bird that is uncommon in New York, but occasionally nests in Prospect Park. I walked more in the woods since the app heard the bird better than me. Of course, I was about to sit under a tree from which the little green bird sang an emphatic “pwee-tseet!”
Merlin Bird ID is more than just a sound identification app, though; it is the result of tens of thousands of bird observers and citizen scientists who have submitted more than a million bird audio recordings to Cornell’s Macaulay Library through the eBird application in just the last few years. Given the volume of data, Weber and Macaulay Library research engineer Grant Van Horn, plus other members of the Cornell Ornithology Laboratory, wondered last summer what it might take to create an identifying feature of the bird song Merlin Bird ID application.
Sound identification is, in fact, a problem of image recognition, explains Van Horn. Caltech and Cornell Tech engineers had already implemented a neural image recognition network toolkit for birds using photos from the Macaulay Library to create the Merlin Photo ID feature. Sound ID converts the audio into spectrogram images, processes them, and then traditional computer vision tools compare these spectrograms to existing bird recording spectrograms.
Crucial to the identification process is a robust training data set — which needed the help of citizen scientists, Weber explains. Like my Bicknell thrush recordings, the Macaulay Library recordings often have several species singing in the background. A team of volunteer annotators traversed the spectrogram formation group from more than 400 North American bird species, drawing boxes around and labeling the sounds of each individual species. The result was a data set with about 250,000 annotations, each box corresponding to a single species. Users of the app upload a file or record the birds live, and the app returns every bird it hears for every three seconds of audio. The team also tweaked the algorithm on a wide variety of background noises, including Google’s vast group of AudioSet data, so that the app was aware of what sounds like birds.
There are other high-quality apps that identify bird-of-fact songs, Cornell’s Ornithology Lab, along with Chemnitz University of Technology, also runs the BirdNET Sound ID app. However, those apps have slightly different purposes: BirdNET serves primarily as a search tool for scientists, while Merlin is instead a science-powered citizen bird identification app that also includes photos and Q + A identification, a guide to integrated field, and data from the eBird city science database of sightings, sounds and images of birds. Data from eBird also helps power Merlin Sound and Photo ID features; they rely on the records of citizen scientists of nearby birds to make more precise recommendations.
There’s a lot of room for Merlin’s Sound ID to grow. There are 10,000 birds, and the app only recognizes 400 of them now. Short circuits pose a challenge, since they can look extremely similar between species, while the application could confuse certain songs at low frequency with the background noise. But as the data set improves, so does the machine learning algorithm and application capabilities.
Van Horn was excited about the potential for the dataset and machine learning model. He plans to use the model in other areas of Cornel’s Ornithology Lab, such as in bird chambers with a constant flow of audio. Weber said that maybe they can use the model to say what the birds are flying over the cities during the peak of bird migration, maybe they can use the model to recognize even the videos of the birds. Van Horn also told me that he thinks about prejudice and other ethical issues of machine learning, and let me know that this algorithm is intended only for wildlife, it was created using only data that l users have agreed to give Cornell via eBird, and it works on the user’s phone. without returning data to Cornell.
The fact that there is a sound identification feature in one of the most popular bird identification apps will be good news for many birders, and after trying it, I can confidently say that it works decently. Experienced birders may also find that their ears are a little more accurate than the app, but, at least for me, the tool was a welcome addition to my bird identification toolkit.