Bird ID Trivia: Merlin vs Birders
[Alli Smith] All right. Welcome, everybody, to tonight’s webinar from the Cornell Lab of Ornithology. My name is Alli. I’m with the Cornell Lab. And I will be hosting tonight’s trivia game. Over the next hour, our expert birder here and all of you in the audience watching will be competing against the Merlin app in a bird trivia game. But before we begin, I have a few quick announcements that I need to make.
So tonight’s webinar is hosted from Ithaca, New York. I want to read a statement acknowledging the Indigenous people as the original inhabitants of this area. Cornell University is located on the traditional homelands of the Gayogohó:no’ or the Cayuga Nation.
The Gayogohó:no’ are members of the Haudenosaunee Confederacy, an alliance of six sovereign nations with a historic and contemporary presence on this land. The confederacy precedes the establishment of Cornell University, New York State, and the United States of America. We acknowledge the painful history of the Gayogohó:no’ dispossession and honor the ongoing connection of Gayogohó:no’ people, past and present, to these lands and waters.
Next up, I have a quick couple of tech notes for our audiences here on Zoom and on YouTube. So closed captioning is available on Zoom. If you’d like to turn captions on or off, you can click the captions button that’s at the bottom of your screen. And if you don’t see a caption button, you can click the three dots that say More and you should see it there.
If you’re watching on YouTube, you can click the CC button at the bottom of the video to turn on the captions there. And one other note about tonight. We have a game show absolutely packed with trivia coming up so we unfortunately won’t have time to answer questions from the audience. If you need tech support during the webinar, please do message us in the chat. We will be monitoring for that.
And if you have a question about Merlin or how to use the app, we have a fantastic Help Center with a lot of answers to common questions. And in the Help Center, you can also submit a message to us there. So we’ll put a link in the chat right now just in case anybody needs that. And that’s it for notes so let’s get started.
So can we have our bird expert and Merlin expert please turn on their cameras? Hello. Hello. All right. So welcome everybody to the Merlin game show. So the game is simple. We will have 10 rounds where I’m going to share a photo of a bird or play a recording of a bird song. And whoever identifies the most birds correctly will win.
So with us today we have Ian Davies from the eBird team who’s going to be competing against the Merlin app to see who can identify the most birds correctly. Ian has been birding for most of his life and loves to share his passion for birds with others in his work as a project leader for eBird. So Ian, welcome to the game show. Do you want to say a little bit about yourself and tell us what your strategy is for taking on Merlin today?
[Ian Davies] Yeah. Thanks for having me. I’m excited to be here today and try to represent birders in the competition against Merlin. So I’ve been birding for almost 20 years. And I’m just excited to see whatever is going to be thrown at us today, whether it’s chickens, or warblers, or hawks, or who knows. We’re here for it. So looking forward to it.
[Alli Smith] Awesome. So tonight Ian is going to be competing against the Merlin Bird ID app, which was developed by us here at the Cornell Lab of Ornithology. And Merlin uses machine learning technology to identify birds in photos or in sound recordings.
So also with us today is Sam Heinrich who is a machine learning engineer with the Merlin app. And he’s going to be sharing a little bit more about how Merlin works behind the scenes throughout the trivia game tonight. So Sam, how do you think Merlin is going to perform today?
[Sam Heinrich] Yeah. I’m excited to see how Merlin does on these ID challenges. Our photo and sound ID algorithms are heavily tested to ensure a pretty high level of accuracy but they have limitations and make mistakes so it’ll be really exciting to see how they stack up against an expert birder like Ian.
[Alli Smith] Awesome. I can’t wait to see how everybody does. So we’re almost ready to begin but first I need to share the official rules so I’m going to share my screen here for a second.
All right. Can we see? Excellent. All right. So in this game, we’re going to be testing the bird identification skills of our expert Ian, of the Merlin app, and of all of you in the audience. So all of the birds in tonight’s quiz will be North American species that can be found in the United States or in Canada.
So here’s how it’s going to work. We will have 10 rounds of questions. In each round, I’m going to show either a photo of a bird or I will play a recording of a bird song. Merlin and Ian will each have about 30 seconds to decide on an answer. And all of you in the audience, on Zoom, and on YouTube will also have 30 seconds to choose your answer from a multiple choice poll that will pop up on your screen.
After the 30 seconds, we’ll close the poll, Ian and Merlin will share their answers, and we’ll see who got it correct. And then at the end of the 10 rounds, the contestant with the most points will be declared the winner. So let’s begin.
The first half of the game, the first five questions, are going to be based on photo ID. So identifying a bird visually is a really important skill that Ian here uses every time that he goes out birding. And when he finds a bird, he’s looking at the bird to look at the size and the shape of the bird along with colors or patterns or the length of the bird’s bill or other visual field marks that he can use to identify it. But Merlin is a little bit different. Sam, could you explain briefly how Merlin identifies birds in photos?
[Sam Heinrich] Yeah, absolutely. So Merlin’s ability to identify photos really starts with the Macaulay Library, which is the Cornell Lab’s state of the art digital archive of natural history photos, audio, and videos. So we had volunteers draw boxes around hundreds of thousands of photos of birds and identify which birds were in those photos. Then researchers used what we call a convolutional neural network, which is a machine learning model that can identify birds in photographs. And the way it does this is by extracting important features such as size, shape, color. And it takes those from an image and it learns to recognize patterns in those features that correspond to a particular species.
So it’s also particularly useful for bird identification because it’s able to recognize two different types of features, one of which is fine-grained and the other is coarse-grained. As you might guess, coarse-grained means like larger scale features. You might think of where there is red on a cardinal, which is everywhere. And fine-grained are smaller features, more detailed. Think about sparrow identification where really subtle color and plumage characteristics can differentiate between two species.
[Alli Smith] That’s super cool. That sounds really similar to how I might identify a bird that I don’t know actually, like using my first overall impression of the size or bold color and then combining that with the finer details to get a really exact identification. That’s super cool.
So let’s jump in with our first question. So for these questions, I’m going to set the scene where you’re birding. And then I’m going to show the photo of the bird. And then we’ll start the poll. And you’ll have 30 seconds to answer.
So question one. You are birding at a park in Ithaca, New York in the spring. And you notice some birds walking around on the grassy lawn.
So here’s the photo. So we’ll take a second to look at this and then you’ll have 30 seconds to answer in the poll that should pop up on your screen. So starting the poll about now. If you all want to start to answer, I will start a timer for 30 seconds. Take a good look at this bird here.
Right. Looks like most the audience is agreeing on one answer. Got about 5 seconds left. And time is up. All right. Ian, what species do you think this is?
[Ian Davies] So this looks like a very familiar friend to many lawns throughout North America. And this is an American robin to my eyes. And you can actually even tell it’s a male American robin because of that really dark black head contrasting with the nice brick red under parts.
[Alli Smith] All right. All right. The audience agrees with you there. We had 876 votes for American robin. And let’s take a look and see what Merlin has to say. We have American robin. And congratulations. Everybody got it right. We all got a point. So American robins can be found across most of North America. And I’m sure this is a really familiar and friendly sight to a lot of people here in the audience. But Ian, what specifically about this bird makes you think American robin?
[Ian Davies] I think for something that is so familiar like this bird, a lot of it’s almost like recognizing the face of a friend when you see them. But really that combination of that red below, that bright yellow beak, and that black head with the white around the eyes there, gray back, all of that and just hopping on a grassy lawn is about as classic as it gets.
[Alli Smith] Awesome. So Sam, did Merlin also find this to be an easy bird to identify?
[Sam Heinrich] Yeah. Images like this are much easier for a machine learning model to identify, and for similar reasons to a human might find it pretty easy. The image is in-focus. It’s well lit. And it’s a really common species so Merlin has probably seen a bunch of these. And it probably picked up on the same features that Ian mentioned, the location of red on the under parts, the black on the head, and the gray on the back.
[Alli Smith] Awesome. All right. So everyone’s starting off with a strong one point. That was an easy startup warm-up question but let’s make it a little bit more complicated. So question number two, we’re going to take a trip over to Toronto up in Canada.
And you are birding in a park. And you see a small brown bird hopping around in a brushy grassy area. So here is a photo. So we’ll take a look at this. And we’ll start the poll. And you’ll have 30 seconds to answer. What species is this?
Ooh. The audience is pretty evenly divided on this species here. Little brown birds can definitely be tricky. Got about 5 more seconds. And time is up. We’ll end the poll. All right, Ian. What species do you think this little guy is?
[Ian Davies] This certainly is not an American robin. This is a fun sparrow. So I see this. I see brown, streaky, kind of small, low to the ground. I start thinking sparrow. And when I look at this, one of the things that jumps out is the fine streaking below with that kind of golden buffy wash to it and just beautiful intricate patterning throughout. And that ultimately points to Lincoln’s sparrow.
[Alli Smith] All right. The audience is agreeing with you. The audience is pretty split on this bird but 40% of the audience also chose Lincoln’s sparrow with closely followed by chipping and song sparrow. And these do all look very similar.
So let’s see what Merlin has to say. Merlin’s agreeing with you, Ian. It is a Lincoln’s sparrow. And congratulations. Everyone gets a point. We’ll count that 40% from the audience as a win. So Sam, how does Merlin handle birds that look really similar?
[Sam Heinrich] Yeah. So sparrows are a real tough one. And you know, like Ian mentioned, they’re all about these really subtle, fine-grained identification features. So Ian mentioned the buffy coloration and that thin streaking. These are color and pattern features which Merlin is quite good at picking out.
And Merlin’s also been trained on an enormous quantity of photos that have been annotated by birders and identified to species. So it’s seen hundreds of images of Lincoln’s sparrows before. And it’s been trained to recognize the difference between Lincoln’s sparrow and all of these other similar sparrows that you’re seeing on your screen right now.
[Alli Smith] That’s pretty cool. All right. So we’ll continue in the tricky bird category with our next question here. So taking a trip all the way from Toronto to San Francisco, you are on a nice walk near the water in the winter in San Francisco.
And you notice a group of shorebirds roosting on these rocks over here by the water. And in that group of shorebirds, there’s one bird that is standing out to you in particular. And it’s this little guy up here on top. So take a look at this bird and we’ll start the poll again. But what species is this bird on top?
Ooh. The audience is very evenly divided on this one again. Good tricky shorebird identification time, especially in the winter when they’re not in breeding plumage and they can all look very, very similar. Definitely a challenge for a lot of birders and for me. All right. We’ll give you 5 more seconds.
And we’ll say time is up. We can end the poll. All right, Ian. What is this absolutely delightful shorebird on top?
[Ian Davies] I love this double-decker shorebird we got going on here. It’s fantastic. The top shorebird is a dunlin to my eyes.
[Alli Smith] What makes you say dunlin here?
[Ian Davies] So it’s a smaller shorebird as you can see in direct comparison to the larger one below it. It’s really brownish overall. So the overall kind of vibes above are brownish. Below it’s pretty clean, pretty white, a little bit of markings along the side of the breast, and then this pretty long bill, longer than its head, that curves down a little bit at the tip.
And many of the other kind of smaller similar shorebirds like semipalmated sandpiper or western sandpiper have a different combination, a different background color, they have more gray usually, not as brown, and smaller, shorter beaks, often more streaking along the sides as well.
[Alli Smith] All right. And this is a controversial one in the audience poll here. So the audience voted for western sandpiper overall. 41% chose western sandpiper. 26% did agree with you, though, on dunlin. And then we had a fair number of folks choosing sanderling or semipalmated sandpiper, too.
Merlin thinks this is a dunlin. And the official correct answer is a dunlin. Nice work, Ian and Merlin. So Sam, these birds, even though they have a slightly different shape, they have really, really similar colored feathers. Like if you were just looking at the color, these two species look almost identical. How does Merlin know that this one on top is a dunlin?
[Sam Heinrich] Yeah. That’s a great, great question. So in certain circumstances, the color of the plumage isn’t a super useful identification mark. And it’s the shape that you really have to look for.
And so if you’ll remember, Merlin is– the actual model that runs photo ID is a convolutional neural network. And one of the features that convolutional neural networks can detect are edges. So they can determine the outline of a bird, the actual shape of an outline, from an image of a bird.
And so when the model is learning, it can learn that certain outlines are really distinctive for certain species and can learn to identify those species by their outlines. So this is super helpful for shorebirds like the ones you’re looking at here on your screen are really good to look at the outline of the bird. And Merlin’s able to do that as well.
[Alli Smith] That sounds really similar to what Ian was saying about some species you can see here have very different like head lengths to bill length ratios. And it really sounds like that’s what Merlin is looking at too. That’s super cool.
All right. We’re going to move on to question number four. So let’s say you’re birding in Cape May, New Jersey, in the winter. And it’s a little bit windy out. And you’re looking out into the Bay. And you see a bird swimming out there. And all the photos that we’ve used so far are really beautiful and very high quality, nice and crisp. You can see the bird really big.
But when you’re out birding that’s not always the reality. Sometimes you do not get a good look at the bird. Sometimes you get a really terrible blurry photo or you only see a piece of the bird, especially if there’s like a warbler in the tree. Sometimes you don’t get a great look. So for this question, let’s say this is the best look that you’ve got. So take a look at this for a second. We’ll turn on the poll. And we’ll give you 30 seconds to try to figure out what this bird is swimming in the bay is.
Winter waterfowl can be certainly very tricky, especially on stormy days. But it looks like the audience is agreeing on for the most part one species here. So we’ll give you 5 more seconds. And time is up. Close the poll.
All right, Ian. I’m sure you’ve had a lot of birding days like this where even though it’s stormy and rainy, you’re going to go out anyway. And you’re going to see some cool birds. And I’m sure you’ve had really awful looks at some birds like this. So what is speaking to you about this picture? What species do you think this is?
[Ian Davies] I think somebody forgot to clean their camera lens before taking the photo. No. What we’re seeing here is definitely just a classic birding view, a lot of ocean birding especially in winter. So the first thing with this I’m looking for is trying to figure out what I’m looking at. And that here appears to me to be a bird facing left.
You can see the head, its beak coming out, and then a little bit of its back. And you’re getting basically everything’s dark except for a white patch on the side of the head and a little bit of white back on the body. And this little round cute head with the little short beak and that white patch on the side of the head to me says bufflehead and specifically a female bufflehead.
[Alli Smith] All right. And the audience is overwhelmingly agreeing with you on this one. We had 55% of the audience said bufflehead. And Merlin says not bufflehead Merlin thinks this is a Brant goose. And I can see that, that short bill, maybe a little bit of white near the neck. It is certainly similar in this photo to a Brant but the correct answer is bufflehead. So Sam, this is interesting. Merlin did not get this one right. Do you have an idea of why that might be?
[Sam Heinrich] Yeah. This is a super interesting one. And it might surprise some people that Merlin is missing it. But Merlin depends heavily on the quality of the photo so some of these really blurry ones and when the bird is obscured are going to be really challenging for the model to get because the model is looking for these really coarse-grained features and it’s really hard to get those when you can’t see the whole bird, and also just the blur on the edge of the bird will make it harder to pick out edges, determine shape, and really identify where those relevant features are. So like you can see here, like Merlin knows what a bufflehead looks like but it has more trouble when you have a worse photo and an obscure view of the bird.
[Alli Smith] Yeah. Yeah. No. That makes perfect sense. I also have trouble when I have a bad view of the bird. But like Ian said, even if it is not a great look, there are always some field marks that you can try to use to pick things out. So that’s interesting.
All right. We have one more photo ID question coming up. So we’re going to virtually head back to San Francisco and California. It’s the spring. And you’re birding at a park. And there are some bushes nearby. And you look over at the bush. And at the top of the bush, there’s a really unusual looking bird sitting up there. So I will show the photo. This is the bird you’re seeing. So we’ll take a quick look at this. And we’ll start the poll again. And you’ll have another 30 seconds to answer.
This is definitely a tricky one. This is not a typically colored bird. And in the past when I’ve seen birds that look like this, it really throws me off. It takes a few minutes sometimes to try to figure out what am I looking at here.
And it looks like the audience is pretty divided on what this species might be. But so far there is a winner here in the audience. We’ll give you 5 more seconds. And time is up. All right Ian, what do you think this bird is? And could you also maybe explain a little bit like what is going on with this bird? Why is he solid white?
[Ian Davies] I think this bird is awesome to start. But what we’re seeing here looks to me like a classic case of leucism or albinism, so basically lacking dark pigments in the feathers. And so you’ll see this sometimes. Leucism is basically partial albinism, so where some birds will be patchy white and then others like this will be almost pure white. And so when you have that, you can’t really look at plumage details, colors, or patterns because they’re all gone.
So for this, it’s really all shape. It’s OK, what do we have here, shape, context, habitat. We have a bird perched up on top of a bush, perching pretty upright, nice long tail, a pretty thin beak that curves down a little bit on the top, but pretty straight and thin, and kind of a friendly eye I might say. Kind of looks like that. And to me, the combination of just the shape and the way it’s sitting point to northern mockingbird for me, a familiar friend on top of many a bush you’ll see driving along the side of the road across much of North America.
[Alli Smith] Yeah. And the audience is agreeing with you there. It was pretty split amongst all four options in the multiple choice poll here but 41% did agree with you that this is a northern mockingbird. And we’ll take a look at what Merlin had to say here.
Merlin thinks this is a cattle egret. And this might be a difficult word to identify but it is in fact not a cattle egret. Ian and the audience are correct. It is a northern mockingbird. So Sam, why do you think Merlin suggested cattle egret here?
[Sam Heinrich] Yeah. First I want to say what a cool picture and what a cool bird. But so birds with unusual plumages like this one are going to be a real challenge for Merlin because Merlin relies on having seen images that look like the one it’s being presented with so it’s seen thousands of normal looking northern mockingbirds but it’s probably never seen one that’s entirely white so it’s not going to be able to realize that, oh, this is a weird plumage bird. I should turn off all of my plumage detection features and just focus on shape like a human can. Merlin’s going to look at the plumage, see it’s all white, and find an all-white bird to try to match it with. So it’s seeing that cattle and that’s why you’re seeing cattle egret as a result.
[Alli Smith] That’s super interesting. That totally makes sense. There are so few– especially in North America– so few solid white birds like this. So it does make sense that Merlin might think it’s an egret if it is heavily relying on the white plumage here. So that’s exciting. That brings us to the end of the photo ID round.
So quick score check. We have a two-way tie between Merlin and the audience with four points each. And Ian is in the lead with five out of five correct. So we’ll see how we do in the sound ID round next.
So next up is our sound ID round. So we’ll have five more questions here. So identifying birds by sound can be a really useful skill. Sometimes you can’t see the bird that’s singing so knowing what it sounds like can be really, really useful. And it’s also fun, too. Like I personally find it really fun to go birding just by listening when I’m doing other things like walking my dog in my neighborhood or if I’m gardening.
And learning bird songs can be really tricky. A lot of birds sound very similar or there’s a huge variety of different noises that they can make. So Sam, could you tell us a little bit about how Merlin learns how to identify birds by sound?
[Sam Heinrich] Yeah, absolutely. So really similar to photo ID, sound ID all starts from the Macaulay Library and this collection of bird vocalizations. So first, sound recording enthusiasts will upload their recordings to the Macaulay Library. And then these recordings are turned into what we call a spectrogram, which is an image representation of sound.
And then a group of highly skilled bird experts go in and draw boxes around the bird vocalizations in the spectrogram and identify which birds these correspond to. Then we take a machine learning algorithm and train that algorithm on these boxes and these spectrograms which creates a model that is able to recognize bird vocalizations in the field and sound ID.
[Alli Smith] So it’s basically the same as photo ID. So instead of recognizing birds in a picture, it’s recognizing patterns of gray like scale vocalizations in a spectrogram instead.
[Sam Heinrich] Yeah, exactly.
[Alli Smith] Yeah. So when Merlin is listening to birds, it’s actually looking at the spectrogram that the bird is making.
[Sam Heinrich] Yeah. It’s super interesting. We’re able to use an image classification task but to identify audio. It’s a really neat concept.
[Alli Smith] That’s super cool. So for this round, we’ll have five questions. So just like the last photo ID round, I’m going to set the scene. And then I’m going to play a song or a call from a species twice. So I’ll give you to the end of the second playthrough plus a few seconds afterwards to answer the poll. I know sound can be a little bit more tricky than photo ID so we’ll make sure you have plenty of time here.
So for question number one, you are birding in a forest in Michigan in the summer. And you hear a very, very loud bird calling from up in the trees. So I will play a sound and we’ll have the poll pop-up. And we’ll see if we can figure out what species this is.
[BIRD CALLING]
All right. Playing it again.
[BIRD CALLING]
All right. So we’ll end the poll here. And it looks like the audience is pretty split between two species. Ian, what species do you think this was?
[Ian Davies] To me, this is a pileated woodpecker, which sounds very similar to northern flicker, another common woodpecker. But to me, the pileated has this deep, rich, loud, resonant quality to it that the flicker doesn’t quite have. It’s a little bit flatter and drier throughout that kind of repeated song. And there are also some woodpecker calls in there as well between the songs, kind of some deep, loud, rich notes.
[Alli Smith] Yeah. And Merlin agrees with you here that it’s a pileated woodpecker. And the audience was pretty split between pileated woodpecker and northern flicker, the other species you mentioned. But 55% of the audience agreed with you on pileated woodpecker.
And the correct answer is pileated woodpecker. Nice work, everybody. Everyone gets a point here starting off sound ID round. Starting off strong. So Sam, did Merlin also find this one easy to identify?
[Sam Heinrich] Yeah. I think this was a pretty easy one for Merlin. So a pileated woodpecker is a relatively common bird. And it’s got that vocalization that’s really just between it and flicker. So it’s seen enough of those two species to be able to differentiate them. Additionally, the recording is of quite good quality. The bird’s really loud. There isn’t a lot of background noise to confuse the model so it’s able to key in on the pileated.
[Alli Smith] That makes sense. That’s cool. So excellent work on question one. We’re going to move on to question number two. So let’s make it a little bit more complicated here, a little bit more tricky.
So for question two, you’re birding in a forest in Pennsylvania. And it’s springtime. And you hear birds singing from up in the trees. So we’ll play the sound again. And we’ll play the sound twice so that you all have enough time to figure out what species this is. Actually, this is a very short recording so we’re going to play this one three times. All right. Here we go.
[BIRD CHIRPING]
All right. Short recording. I’ll play it again.
[BIRD CHIRPING]
And one more time. Look at that beautiful spectrogram.
[BIRD CHIRPING]
All right. Common sound in the forests of Pennsylvania and other places in the northeast in the summer. We’ll end the poll. And looks like we have some split opinions here. Ian, what species do you think this might be?
[Ian Davies] Yeah. So this is a fun one. This is a tricky group of songs. So when I hear kind of simple whistles that have these long spaces between them, I immediately think vireo, so small songbirds like warblers, a little bit more robust. And so often you just hear them and you never see them.
So this bird has these whistles spaced apart. And the whistles are very clear to me, very piercing, pretty simple notes. And they have a little bit of like huskiness them sometimes but sometimes they don’t. Sometimes they’re just clear. And so to me when you have simple, very clear, sometimes husky, sometimes not, that’s a blue-headed vireo. All right. And the audience is agreeing with you here. 49% agreed blue-headed vireo here though the next most commonly chosen answer was a Cassin’s vireo.
So let’s see what Merlin has to say. Merlin is agreeing with you that it’s a blue-headed vireo. And the correct answer is blue-headed vireo. Nice work, everybody. Everyone gets another point.
So Sam, this bird sounds very similar to a lot of other species, including the Cassin’s vireo that was just mentioned. How does Merlin tell these two apart?
[Sam Heinrich] Yeah. So these two are extremely similar, both visually and by their songs. And so that’s why it’s super important to have your location on when you’re using Merlin because Merlin will actually take into account your location information.
So the blue-headed and Cassin’s vireo don’t really ever overlap in their range. And so if you know where you are, you kind of know whether you’re hearing a blue-headed or a Cassin’s. And so this sort of information is really important for sound ID to have because it will help direct you to the right species based on your location. So if Merlin didn’t have your location, it might get this question wrong.
[Alli Smith] Yeah. So important takeaway here for the audience. Next time you’re using Merlin, make sure your location is turned on. Make sure you have location permissions enabled for the app so that Merlin knows where you are so that you get the most accurate results possible. So that was a tricky one.
So let’s continue in the tricky category. These are always fun. So question number three. We’re going to head back up north to Minnesota. And you are birding in Minnesota. You’re in a spruce forest in the spring. And you hear a trilling noise coming from in the trees. So I will play the song twice and we’ll have another two rounds of the song for you to answer in the poll that will pop-up.
[BIRD TRILLING]
Strong opinions from the audience here. All right. We’ll play it one more time.
[BIRD TRILLING]
All right. Time is up. We’ll end the poll. Ian, there’s a lot of different species that sound similar to this. What species do you think was this one singing?
[Ian Davies] Yeah. This is one of the tricky trillers is what I always think of them as. There’s few species that just have these kind of even trills of notes. So chipping sparrow, dark-eyed junco, pine warbler, worm-eating warbler are some of the classics for that. And so to me one of the big things I listen for is the kind of quality of the trill. Is it really sharp and kind of mechanical, just very robotic almost, or is it more musical and soft?
Pine warbler is kind of the softest and most musical. Then junco is a little bit less musical, a little bit kind of harsher, and then chipping sparrow is even more harsh, and then worm-eating warbler sounds almost like an insect. It’s so mechanical and robotic. And so to me, the mechanical nature of this and the evenness of it, it didn’t really fade in or out, it just kind of starts and then goes and then stops, sounds like a chipping sparrow to me.
[Alli Smith] All right. The audience is agreeing with you pretty strongly here. 47% of the audience said chipping sparrow followed by dark-eyed junco and worm-eating warbler. So let’s see what Merlin has to say. Merlin is agreeing with you that Merlin thinks this is also a chipping sparrow.
And the correct answer is chipping sparrow. Nice work. Everyone gets another point here. Doing great. So Sam, how does Merlin handle this, the tricky trillers here?
[Sam Heinrich] Yeah. I mean these trillers are really a tough ID challenge for humans and for Merlin. But luckily, we’ve got this expert community of annotators and so we have many labeled recordings of both chipping sparrows, dark-eyed junco, worm-eating warbler, all the trillers.
So the models learned how to differentiate them in most circumstances. But they can have some overlap. And the model can get them a little– can mistake them. So it’s really important to verify your results. So go out and find whatever Merlin is telling you you’re hearing.
[Alli Smith] That’s always good advice. I know sometimes it’s not possible to always go and see it. But that’s always a really great way to confirm what you’re seeing, especially these birds in particular. They all look very, very different. So even if you only get a little glimpse of what’s singing, you might have a really good clue as to what you’ve got.
So moving on, we’re going to head down south to South Carolina to a nice city park in Charleston. And it’s the middle of the summer. And you’re walking in this park. And you hear a really unusual song coming from up in a tree. And the song sounds something like this. So we’ll start the poll again. And we’ll listen to this song twice.
[BIRD CALLING]
All right. I’m going to play one more time.
[BIRD CALLING]
All right. So Ian, this almost sounds like a whole chorus of different species. What is going on here?
[Ian Davies] Yeah so this bird was a blue jay, and an eastern king bird, and a northern cardinal, and a Carolina wren, and an American kestrel, and I’m sure some other things mixed in there as well. So this is a mimic.
And so across North America, we have a set of different mimics. And the best way that I think of, especially in the Eastern US, to tell them apart is basically how many times they’re imitating whatever they’re imitating. So the three in the Eastern US are gray catbird, brown thrasher, and Northern mockingbird.
Great catbird usually just kind of has these little warbles and then throws in a single mimic most of the time, just one little phrase, maybe two. Brown thrasher pretty reliably repeats things twice, something twice, another thing twice, et cetera. This one was doing things three, five, six, seven times in a row. So for me, this points towards a northern mockingbird. And I don’t know if it’s white or not, but it’s a northern mockingbird to me.
[Alli Smith] Let’s see what Merlin has to say. Merlin has a lot to say here. The audience agrees with you. 70% overwhelmingly agreed that it is a northern mockingbird. And Merlin does list northern mockingbird in the list of species here.
So the correct answer is northern mockingbird. You and the audience got it right. Technically, Merlin got it right here, too. I think we can award Merlin a point here. But Sam, what is going on here? Why did Merlin list four different species for this?
[Sam Heinrich] Yeah. So as you might imagine, mimics are a huge challenge for Merlin. And you might have had experiences like this in the field where you’re listening to something and you’re like, oh, that sounds like this but it also sounds like this. Like Ian was saying, it can do like six different species.
And then you’re like, oh, after you hear them a bunch of times, you’re like, oh, I think it’s a mockingbird. You take all that information of listening to the bird for 30 seconds, a minute, and you kind of put it all together to realize that it’s a mockingbird. Merlin doesn’t have that ability. Right. So Merlin looks at 3 second increments of what you’re listening to. And it can only look at those 3 seconds.
So it can be– it’ll hear the Eastern king bird that it does and think, OK, that’s a king bird. And then it’ll wait a bit and then it’ll hear the kestrel and say, OK, it’s the kestrel. But what it can’t do is combine those two results and be like, oh, OK, I think this is actually a mockingbird. So it doesn’t have the opportunity to remember what it heard five, 10 seconds earlier in the recording.
[Alli Smith] That’s super interesting. So like Ian is able to take that whole big picture into account and can listen for the full 30-second recording or for minutes at a time even and can piece that together but Merlin is really just limited to that little 3-second chunk. So that’s super interesting.
And honestly, I’m impressed that Merlin can pick out that many species within the song. It did a better job than I did of figuring out what this bird was mimicking. So that’s pretty cool. I think I think Merlin gets a cool point for that. So we’ll count that as correct for the purposes of the game show.
So we’ve got one final question here. We have sound ID question 5. And this question is going to be a little bit different. So we’ll set the scene again. You are birding in North Dakota in a freshwater marsh. And it’s summer. And the question this time is how many species are singing. So I don’t need identifications.
If you really want to, you can identify the species that are singing here, but I just want a number, somewhere between one and 10. How many different species of birds are singing in this recording? And this is– it is a loud recording so I’m going to play it twice again to make sure that you all have enough time to try to keep track of how many different species are singing here. So we can open up the poll. And we’ll start the sound.
[BIRDS SINGING]
All right. That’s busy. I for sure have a tricky– I have a very hard time with tricky dawn choruses like this so I’m going to play it again so you can all try to count one more time.
[BIRDS SINGING]
All right. That was a very busy recording. I’ll let everyone in the poll take a few more seconds to get their answer in. I’ll let Ian think for a second. And all right. Let’s end the poll. And so Ian, when you are confronted with an absolutely chaotic dawn chorus like this for I guess I have two questions. How many species do you think are singing and then, number two, how do you even begin to start pulling different species out of this?
[Ian Davies] Yeah. This is actually a cool example here. I’ve never had to think about trying to parse out all those songs in such a short time frame before. So I think there were seven or eight but I found myself getting really hung up on the common ones and, frankly, just trying to count on my fingers and remembering which ones I’d counted or not.
So there were definitely some really loud marsh wrens, Wilson’s snipe, red-winged blackbirds, common yellow throat. There was American bittern pumping in the background, duck calling that sounded like a gadwall at the end. There was something kind of cooing a little bit that could have been a distant trumpeter swan maybe but it sounded different. So I might go and I’ll just guess that I missed one so I’m going to say eight for that. But I truly don’t know and I’m really excited to see what Merlin has to say.
[Alli Smith] Yeah. So the audience was pretty torn. 36% picked seven species, about 17% decided eight species sounded good, and 33% decided six. So everyone’s agreeing there’s a lot of birds calling here. This is a really tricky question. So let’s see what Merlin had to say.
Merlin says seven species. And this is the correct answer. There are seven different species calling in this recording. Ian, you actually you got six of the seven species there so congratulations. That was awesome. That’s like way more than I did when I heard this recording for the first time. So that’s a lot of birds. Sam, how does Merlin make sense of this completely chaotic recording here?
[Sam Heinrich] Yeah. So this is one scenario where Merlin has kind of an advantage on humans and it’s actually really cool. So I’m going to dive into it a bit. So it can be challenging for people to focus on multiple vocalizations at once. You’re like trying to figure out is that a yellow throat. Is that also a bittern in the background? Usually, you kind of have to focus on one thing, identify it, move on to the next one.
Merlin has what I like what I like to call an infinite attention span, meaning it can listen for thousands of species at once. And so every time you start up Sound ID, Merlin is listening for the 1,000 plus species that are available in Sound ID. And it’s making a determination on whether it hears each of those 1,000 plus species every single second you’re recording. So for this, it’s heard all these species before. It’s heard them in different backgrounds. And so it’s able to identify them each time that they’re singing.
[Alli Smith] That’s super impressive. I wish I had Merlin’s attention span. I for sure don’t. Yeah. I’m sure like a lot of other birders, I find it very easy to key in on one species but the rest sometimes get lost in the background. So it’s cases like this where I’m really thankful to have Merlin in my pocket and to be a little birding buddy with me or though maybe I need to bring Ian along birding more often.
So all right. That brings us to the end of our game show here. So that was 10 questions. So let’s do our final score check here. We have Merlin with a respectable eight out of 10 questions correct.
We have the audience with nine questions correct and Ian also with nine correct. So Ian and the audience, congratulations, you have tied for first place. You have beaten Merlin in the bird trivia game show. Nice work. Excellent job.
Thank you so much for participating here. Everyone did a really great job. Thank you so much, Sam, for explaining how Merlin is working behind the scenes. Thank you, Ian, for sharing all your ID tips and tricks. I hope everybody learned something new today. I for sure did. And thank you to everybody in the audience for participating.
I do want to leave all of you with some tips for using Merlin while you’re out birding next time. So number one is, of course, always make sure that location is on when you’re using Merlin because without your location, Merlin is rather than just trying to figure out what bird might be singing in your immediate area, it’s considering all of the birds of the world. So if you’re using Merlin with location turned off, you might get some really wacky results so making sure that it’s on we’ll make sure that you are getting the most accurate results possible.
And then if you’re using photo ID like we saw with our bufflehead photo here, sometimes blurry photos don’t go over so well with photo ID. So whenever possible, I know it’s not always easy, especially if all you have on hand is your phone, using clear photos in Photo ID will get you the most accurate results.
And same with Sound ID. Minimizing the amount of background noise when you’re using sound can really, really help. Like you saw how fuzzy some of those recordings might be. If you’re handling your phone, or moving your hands around a lot, or if you’re talking in the background, or near a busy road, sometimes Merlin will have a harder time hearing the birds that are around you. So minimizing background noise can really help boost your sound ID game up a little bit.
So with that, I want to thank everybody again one more time. Thank you, Ian. Thank you, Sam. Thank you everybody for watching. And we will see you next time.
End of transcriptJoin us for a bird ID trivia game and learn some new bird identification tricks! The Cornell Lab’s Merlin Bird ID app is powered by machine learning. Can an app outsmart experienced birders and a live audience? Using bird photos and sounds, we’ll see who can correctly identify the most species. The game involves audience participation, so come ready to play! During the game, we’ll learn how Merlin “decides” on an ID, and panelists will share their tips for sleuthing out tricky birds by sight and sound.