I wish when I uploaded a photo for analysis, and the app identifies my bird, that I could keep MY photo and not the generic one in their database. It's great at building a checklist, but not so much for making it feel like my own.
If you use the companion app called ebird, you can save your checklists. Later you can upload your photos and any audio you have recorded. Those get vetted by regional bird experts and get added to the scientific data. It's fun to go back and relive past hikes and trips viewing your pictures on the ebird.org website!
On the Android app, tap the share button at the top of the recording play screen. This lets you export the audio WAV file. Select some other audio player app to share to, such as Podcast Addict, and use that app's Cast feature once it opens.
I'm friends with some of the researchers on the Sound ID portion of this app! The team's gone to great lengths to make sure the machine learning models and evals are solid.
Under the hood, Sound ID is a great example of how "domain-expert-driven" careful research can give more reliable results than just feeding in data and hoping for the best.
I'm definitely impressed with the sound ID. One time there was a cacophony of birds singing outside my house at varying distances, and the app was able to identify 6 different bird species within 30 seconds. All 6 suggestions seemed reasonable to me.
We sometimes joke it's hallucinating when it detects something unexpected, but often it was being accurate. I saw my first ever male Blackburnian Warbler a couple weeks ago after Merlin picked it up.
I would like to see a version of the app that assigns names, as in 'Fred', 'Bertha', 'Kevin' to given birds, algorithmically. I get the same birds in my garden and I already know what species they are, but, I wouldn't mind knowing if my favourite birds have got new mates, or, if I am down the road, some way from home, to confirm that it is 'Kevin' that I see, as in the same 'Kevin' that frequents my bird bath.
Consistently assigning names to different blackbirds might be tricky, but, for other birds, I am sure that some AI algorithm could do this.
Pass them my thanks. I've been using the Sound ID on dog walks for a few years now and I love it. It's made me much more interested in the birds around me now, and if it tags a new bird that I've not come across before I'll often try to spot it. I think I'm a bird spotter now.
I like how the app shows reference calls/songs for detected birds so I can verify using my own judgement, or to figure out which birds are which when there's a lot of chirping going on.
Someone near my house has a rooster that does his thing every morning. And every morning I look at the app hoping to see him pop up in the ID list but alas, it never IDs him.
Please Let your friends know that roosters are birds too and deserve their recognition as the world's wake up call! Lol
I'd love an API of some sort. I have some birdwatching/tracking ideas that require I can id a bird by sound and Merlin would be the best route if they just let people us it ...
I was really impressed when I took a trip to England and it recognized european birds. At one point it said there was a Rose-ringed Parakeet which seemed totally wrong, but then we saw a bunch of them in Hyde park a couple days later.
The iOS app did seem to lock up when trying to download the region packs though. I should find out where to file a bug!
In general as a bird watcher i’ve been extremely impressed by this tech. I generally trust it.
There’s only a couple of times i’ve been sceptical of it’s id and thats where there’s similar species in the area. Eg. I’m not convinced there really is a purple finch where i live when all i see is house finches all day. But i could be wrong too! It’s proven itself enough that i’m not ready to call it wrong on that one.
The only errors I've seen it make with common birds in the UK are also with finches - specifically with the greenfinch's "at rest" twitter, which it consistently mistakes for a goldfinch.
The two are visually distinctive but, in Merlin's defence, I can't tell them apart by ear either!
It definitely has trouble with similar species sometimes — I've noticed it recently with crows and warblers. But it does a great job generally and direct observation of the bird usually clears up the confusion.
I have a bunch of Blue Jays around my house and it turns out they are so good at imitating a specific hawk species(blanking on its name) that Merlin actually reports it as the hawk! I went and listened to real recordings of that hawks call and I couldn't tell it apart from the Jays imitation call.
Now, I know it's technically possible it was a real id, but Im pretty sure the bald eagle it detected was actually one of the kids down the street running around screaming lol
Surely your local blujays must have heard the hawk somewhere in order to learn its call? It figures that a hawk would occasionally show up in your neighborhood.
Machine identification can still need some manual confirmation, even though this app does a really good job. It’s not a confirmed sighting until you have a visual confirmation.
In this case, while I couldn't see it directly while he was making the call, I had watched him fly to and land in a nearby tree before the call and then shortly after watched it take off. So I can't be 100% sure, but I'd be willing to bet that it was the Jay.
I've been birdwatching (birding?) actively for a few years now, but only this year did I start using these sound identifiers. What a boon it's been! I've already spotted over a dozen new species by sound alone, and also learned to identify some of them by myself.
It really has opened up a whole new venue of enjoying this hobby. At least here, machine learning/AI has a clear, positive impact.
I was visiting Yosemite a year-ish ago and a guide recommended this app and it was fantastic. My kids also loved getting the real-time info on what they were hearing, and trying to spot the various kinds of birds.
Strongly recommend it (though admittedly, I don't use it often in suburbia)
I never thought I would ever actively watch for birds when I hear them. Or that I would be able to say "this is clearly a Wren". Or that I have a my favourite bird (Wren as well btw).
I won Merlin once. I didn't know you could win the game, but I was relaxing in my backyard and suddenly it notified me that I had found the merlin. I was so excited. I hadn't known that a merlin was a kind of bird.
I do wonder how accurate Merlin is. I certainly can't tell the bird calls apart, and I don't usually spot the bird in question, so it could just be lying to me half the time, and I'd probably never notice. But I sure do love the app as an amateur bird-liker.
Fellow amateur bird-liker here. Usually when you ID a bird it’ll provide you with a boatload of other recordings you can compare against if you’re unsure.
You can also use these recordings to annoy birds. Play their species' song and the bird will think there's a competitor nearby, and sometimes fly close to confront the interloper.
Needless to say, don't do this too much, but it can be useful for getting a visual on certain birds (like warblers).
I love this app!! This and iSeek (plant/animal identification via photos) are a blast. Lived in the jungle in souteast Mexico the last couple years and I'm constantly reaching for these.
We have a Mockingbird in our backyard (or more precisely, we live in his/her territory) that impersonates a Gila woodpecker. We were able to record it. Playing back the video and using Bird ID actually shows it as a Gila Woodpecker.
Oh, that's not my experience at all! I downloaded this app to see how many distinct songs a mockingbird was singing one time (it was a lot) and the app just came back: "Mockingbird". I was both amazed and disappointed!
I have a Blue Jay that began imitating a Cooper's Hawk after one started showing up in the neighborhood. In its early attempts at imitation, it was still recognized as a Blue Jay, but the Blue Jay has improved and it now comes up as a Cooper's Hawk! I watch it sing, so I know this is the Blue Jay, not the hawk.
Came here for this. My wife is a big Merlin fan, but I'm pretty hard of hearing and much more visually-orientated when it comes to the natural world. INat is a fantastic project, I've learned so much through it.
Note that the recent iOS update seems to have introduced a bug whereby even after stopping recording, the app in the background will continue to use SIGNIFICANT battery life, so I've needed to start force-closing the app after using it.
I identified a https://en.wikipedia.org/wiki/Red-whiskered_bulbul in our garden thanks to this amazing app! Me and my wife love birding and at the end of the year we compare our scores, it is playful entertainment. And we are introducing our daughter to it, it is so cute when she identifies one!
Merlin is far less capable in developing countries either because there’s less available crowdsourced data, or it’s intentionally suppressed to reduce its use by unskilled poachers
In particular, it makes nocmig drastically more approachable. Be warned, though, that the false positive rate is extremely high (which is to be expected).
It's very nice, I prefer it to Birdnet-pi. It's been running very well for a few months now. It's cool to see the daily and seasonal changes in the detections.
Fantastic app. My wife and I are gradually becoming able to recognise more and more birds by their call. This is great because the human ear is better than the app at rejecting environmental noise - so we can spot birds by their call even close to a noisy road when the app cannot.
The app, at least on my Pixel 6, struggles with very high frequency calls - e.g. long-tailed tits.
I find it great for familiar birds that only make certain calls when they are less visible and others out in the open. Let's you connect the variety of sounds throughout the day to specific species.
I feel like we share interests and location. The long-tailed tits in Wimbledon present no trouble for Merlin my Samsung phone, but I've never seen a nightingale round here.
> Sound ID is trained on audio recordings that are first converted to visual representations (spectrograms), then analyzed using computer vision tools similar to those that power Photo ID.
Yeah, the spectrogram scrolling by at the top isn't just a cute gimmick, that's actually how the recognition works...
When working in a linguistics lab as an undergraduate long ago, we looked at spectrograms to identify sounds (specifically places of articulation) as much as listened to recordings.
So it makes some sense to build a model on them rather than some other representation of the sound.
I had friends in MIT's computational linguistics group back in the 1980s who did a casual get together where they'd take turns handing out spectrograms of human speech and for the rest of the group to try and interpret. Apparently this started with some noted researcher asserting that you couldn't interpret a spectrogram faster than it was originally spoken, and one of them decided to learn them well enough to disprove it by example, which turned out to be easy but inspired them to continue by trying to make more challenging ones - culminating in getting John Moschitta Jr. to record for them :-)
Me too. I was out hiking and ran into an old guy I know. He is 90 years old and he's a bird biologist and a naturalist by vocation. I changed my plans for the day and hiked with him for a few hours. He talked quite a bit about birds and ecology and it sparked my interest. I figured someone must've made an app to id bird calls, and I found Merlin in the Google store. Now I've been walking around trying to id every bird I hear. That was a few days ago, so seeing it here on HN is definitely a Baader–Meinhof moment.
Wonderful app! Collecting birds is great fun and another fun use is to try and fool it by making my own bird noises and see if I can pass. Red-tailed hawk is my speciality. I can fool it ≈50% of the time once I’m warmed up.
I’ve used this app for a while, it’s really good and I’d highly recommend it if you want to learn more about the birds where you live in an accessible way.
220 comments
[ 3.4 ms ] story [ 102 ms ] threadThe only one I've seen it misidentify is a starling but apparently those are excellent mimics so I don't blame the app.
Under the hood, Sound ID is a great example of how "domain-expert-driven" careful research can give more reliable results than just feeding in data and hoping for the best.
We had a brown thrasher stringing together a long series of mimicked calls that got labeled correctly as a brown thrasher.
I don’t even know how you’d approach the ML for that seeing how unique each song is to each bird
The waveform was too thick and looked funky which is why I tracked down the bird doing it. (I Was at the Ohio bird sanctuary)
Consistently assigning names to different blackbirds might be tricky, but, for other birds, I am sure that some AI algorithm could do this.
Please Let your friends know that roosters are birds too and deserve their recognition as the world's wake up call! Lol
https://birdnet.cornell.edu/api/
https://birdnet.cornell.edu/
Merlin is trained on the same source as BirdNet.
You can also use the iNaturalist API for visual ids, for what it's worth.
https://www.inaturalist.org
The iOS app did seem to lock up when trying to download the region packs though. I should find out where to file a bug!
There’s only a couple of times i’ve been sceptical of it’s id and thats where there’s similar species in the area. Eg. I’m not convinced there really is a purple finch where i live when all i see is house finches all day. But i could be wrong too! It’s proven itself enough that i’m not ready to call it wrong on that one.
The two are visually distinctive but, in Merlin's defence, I can't tell them apart by ear either!
Now, I know it's technically possible it was a real id, but Im pretty sure the bald eagle it detected was actually one of the kids down the street running around screaming lol
Machine identification can still need some manual confirmation, even though this app does a really good job. It’s not a confirmed sighting until you have a visual confirmation.
In this case, while I couldn't see it directly while he was making the call, I had watched him fly to and land in a nearby tree before the call and then shortly after watched it take off. So I can't be 100% sure, but I'd be willing to bet that it was the Jay.
It really has opened up a whole new venue of enjoying this hobby. At least here, machine learning/AI has a clear, positive impact.
Strongly recommend it (though admittedly, I don't use it often in suburbia)
I never thought I would ever actively watch for birds when I hear them. Or that I would be able to say "this is clearly a Wren". Or that I have a my favourite bird (Wren as well btw).
I do wonder how accurate Merlin is. I certainly can't tell the bird calls apart, and I don't usually spot the bird in question, so it could just be lying to me half the time, and I'd probably never notice. But I sure do love the app as an amateur bird-liker.
Needless to say, don't do this too much, but it can be useful for getting a visual on certain birds (like warblers).
I might check out Merlin specifically just for the ID wizard, for raptors which are pain to photograph and are best IDd by shilouette.
[1] https://www.allaboutbirds.org/guide/merlin/id
[0] https://www.birdweather.com/birdnetpi
The app, at least on my Pixel 6, struggles with very high frequency calls - e.g. long-tailed tits.
My best spot yet? A nightingale in Wimbledon.
> Sound ID is trained on audio recordings that are first converted to visual representations (spectrograms), then analyzed using computer vision tools similar to those that power Photo ID.
Yeah, the spectrogram scrolling by at the top isn't just a cute gimmick, that's actually how the recognition works...
When working in a linguistics lab as an undergraduate long ago, we looked at spectrograms to identify sounds (specifically places of articulation) as much as listened to recordings.
So it makes some sense to build a model on them rather than some other representation of the sound.