I downloaded the cornell Merlin Id and found it inaccurate, I live in the midlands of england. On setting the programme in my garden it found a Kingfisher where there is no water and on a Pied Flycatcher which is scarce in my location. On other occasions it showed Long Eared Owl and water Rail, this app cannot be trusted.
-
Welcome to BirdForum, the internet's largest birding community with thousands of members from all over the world. The forums are dedicated to wild birds, birding, binoculars and equipment and all that goes with it.
Please register for an account to take part in the discussions in the forum, post your pictures in the gallery and more.
cornell merlin id (1 Viewer)
- Thread starter greater
- Start date
No but it's a useful starting point. You clearly know enough to know when it's ids are unlikely. I agree it would be "dangerous" for a novice user
Jon.Bryant
Well-known member
I have found it pretty good in general, but it does make blunders, so should be treated as a pointer to ID rather than gospel. I think it works better for longer songs, but is worse on short sounds and calls.
I presume that this is because the pattern matching algorithms works best on longer extracts, but also probably because there is a greater dataset for bird songs as a reference point for the analysis. Assuming the latter is true, then the answer to making the software more accurate, is for us all to spend more time recording all the sounds we can, and to make these available online and to Merlin.
For your example of Pied Flycatcher there are 434 recordings on the Macaulay Library, but the majority are of song rather than call - I presume your erroneous ID was from a call? In order to generate more accuracy there needs to be a good dataset of reasonable quality recordings of Pied Flycatcher, but also of everything else that may sound similar to it.
I think the expectation is that the software will in fact improve in accuracy over time as hopefully the amount and breadth of reference data increases.
I presume that this is because the pattern matching algorithms works best on longer extracts, but also probably because there is a greater dataset for bird songs as a reference point for the analysis. Assuming the latter is true, then the answer to making the software more accurate, is for us all to spend more time recording all the sounds we can, and to make these available online and to Merlin.
For your example of Pied Flycatcher there are 434 recordings on the Macaulay Library, but the majority are of song rather than call - I presume your erroneous ID was from a call? In order to generate more accuracy there needs to be a good dataset of reasonable quality recordings of Pied Flycatcher, but also of everything else that may sound similar to it.
I think the expectation is that the software will in fact improve in accuracy over time as hopefully the amount and breadth of reference data increases.
Well my internal one certainly does...
Steve Lister
Senior Birder, ex County Recorder, Garden Moths.
I have used the sound ID app mainly in demonstrating it to other people. Almost every time it comes up with species that none of us heard and did not expect to hear. Kingfisher and Tawny Owl I put down to nearby steam trains. Pied Flycatcher, Firecrest, Redwing, Red Crossbill, Woodlark and American Goldfinch (at Rutland Water!) have all come up rather unexpectedly. The Crossbill might have been some hen chicks cheeping but no idea where the others came from.
As stated above, it should improve as more reference calls are added to the database. At the moment I don't think it is recommendable. A friend who uses it more than me says the same - it has come up with both Icterine Warbler and Bluethroat several times for him.
As stated above, it should improve as more reference calls are added to the database. At the moment I don't think it is recommendable. A friend who uses it more than me says the same - it has come up with both Icterine Warbler and Bluethroat several times for him.
Jon.Bryant
Well-known member
This thread is rather similar to Is anyone here using Merlin? It is revolutionizing my birding
Personally, I would recommend Merlin to others - see my comments in the other thread regarding using the app on the Attwater autoroute in Texas.
Perhaps the software works better in the states, but to be honest I found the software a bit of a life saver there - yes I should have done my homework and learnt the songs prior to my holiday, but I didn't. Merlin helped before I got a bit more familiar with the songs - there were quite a few times when the software made an ID, and I was able to say 'just another [Cardinal, Bewick's Wren etc. etc.], move on.
I would say use it but don't rely on it - I wouldn't recommend spending hours searching for a bird that the app IDs [the unlikely Bluethroat or Icterine Warbler], particularly if you can't make out the sound yourself. As I say it is just as useful at proving the familiar - last year I had a European Reed Warbler doing subsong from a hedge near the front of my house (not a reed or even any water in sight!). The bird was singing from deep cover and I only got glimpses of it. Timing wise it was late May, so I started to think anything was possible - perhaps it wasn't even an Accrocephalus? It was reassuring (though disappointing) that BirdNet confirmed my initial assuming that it was a European Reed Warbler. (BirdNet and Merlin are both Cornell, and I think they probably use the same analysis engine).
I wouldn't ID a bird on just another birdwatchers say so - I would want to ID the bird myself. I think you should treat Merlin similarly and confirm (or not) its suggestions.
Personally, I would recommend Merlin to others - see my comments in the other thread regarding using the app on the Attwater autoroute in Texas.
Perhaps the software works better in the states, but to be honest I found the software a bit of a life saver there - yes I should have done my homework and learnt the songs prior to my holiday, but I didn't. Merlin helped before I got a bit more familiar with the songs - there were quite a few times when the software made an ID, and I was able to say 'just another [Cardinal, Bewick's Wren etc. etc.], move on.
I would say use it but don't rely on it - I wouldn't recommend spending hours searching for a bird that the app IDs [the unlikely Bluethroat or Icterine Warbler], particularly if you can't make out the sound yourself. As I say it is just as useful at proving the familiar - last year I had a European Reed Warbler doing subsong from a hedge near the front of my house (not a reed or even any water in sight!). The bird was singing from deep cover and I only got glimpses of it. Timing wise it was late May, so I started to think anything was possible - perhaps it wasn't even an Accrocephalus? It was reassuring (though disappointing) that BirdNet confirmed my initial assuming that it was a European Reed Warbler. (BirdNet and Merlin are both Cornell, and I think they probably use the same analysis engine).
I wouldn't ID a bird on just another birdwatchers say so - I would want to ID the bird myself. I think you should treat Merlin similarly and confirm (or not) its suggestions.
Last edited:
Stonefaction
Dundee Birding....(target 150 in 2024).
For potentially 'dodgy' records, once you've finished recording you can select the questionable ID from the list and the recording will jump to just before where it 'heard' it and you can play it back and listen to it yourself as many times as you like, (and if you're experienced enough you can work out what it actually was, or use sonagram etc to help get an ID later). I've found through using the app that my brain seems to filter out some calls (House Sparrow chirps to a large degree) and that when I listen back to the recording I realise that I did actually just miss them.
faustdownunder
New member
Hi
Just starting with Merlin ID using the Sound ID function. I am in a relatively quiet suburb in Melbourne, Australia. Whenever I use the Sound ID function it gets the "Eurasian Blackbird". While this bird may be there, I could clearly hear magpies, parrots, mynas, and other birds. I don't know why this is happening. I checked my data base and saw that all the usual Australian birds are there.
Maybe an experienced user can explain. I cannot exclude a mistake on my behalf.
thanks
Dieter
Just starting with Merlin ID using the Sound ID function. I am in a relatively quiet suburb in Melbourne, Australia. Whenever I use the Sound ID function it gets the "Eurasian Blackbird". While this bird may be there, I could clearly hear magpies, parrots, mynas, and other birds. I don't know why this is happening. I checked my data base and saw that all the usual Australian birds are there.
Maybe an experienced user can explain. I cannot exclude a mistake on my behalf.
thanks
Dieter
DMW
Well-known member
As others have noted, Merlin is a work-in-progress and is less "competent" outside the US, and is best viewed currently as a suggestion rather than an answer. As the dataset and algorithms improve, so will it's usefulness. It's probably also worth noting that species with mimicry or that just have a very wide and individually variable repertoire are always likely to be problematic. As a human listener, we have "3D stereo" hearing where Merlin is dealing with 2D mono, which I suppose makes it harder to isolate individual birds within the total soundscape.
faustdownunder
New member
Understood. Based on my professional exposure to signal processing, I wonder whether "machine learning" would be easier if the recorded time domain data (waveform) were transformed into frequency domain (spectrum). Furthermore, spectral density consideration may perhaps also yield better identification. Meanwhile I found another local software (App) which seems to work well. thanks
Merlin only works for North America and Europe (to a lesser extent). Another app, BirdNET, recognises roughly 3/2 times more species, but it's still confined to North America, Europe and Northern Africa. I don't know of any app working in Australia; glad you found one, though.
faustdownunder
New member
Yes, thanks. I can now identify birds quite fast and reliably. The app is called "Smart Bird ID". It was apparently created by a single worker.
I have found Merlin quite useful, but on occasions my Samsung galaxy phone isn't picking up some sounds, particularly low register (eg Corvids)
Is there a reasonably priced plug in or Bluetooth microphone that could be clipped on jacket with the phone safe in pocket?
Is there a reasonably priced plug in or Bluetooth microphone that could be clipped on jacket with the phone safe in pocket?
LittleBitOfBreadNoCheese
Well-known member
That birdNET info is a bit out of date. Certainly they have been adding South American and Asian species. I can't find the true total but some 24h sessions have hit 1300 so it must be at least that. They have a Twitter account you can follow to keep up for anyone who is interested. They tweet each new addition to the database.
Last edited:
Thanks for the heads-up. It feels quite special to see both BirdNET and Merlin (with all the new Asian bird packs) develop before my eyes. (At the same time BirdNET still has no clue in the face of the staple male Indian Peafowl, but I suppose we'll get to that point someday.)
GreenRooky
Well-known member
Anyone use any kind of mic with their phone when using these apps. I generally just hold it in my hand as I walk around but if it picks up something unusual and I go to listen back all I hear is wind noise generally, not sure if there is anyway to get a better result
Direct the mic towards the bird, not yourself. Nothing more--as far as I know--and I don't plug anything in. Distant sounds are not easy to pick up for my phone as well.
songbirder
Member
For what it's worth, if anyone has a recording from the Merlin app where they are certain the ID that Merlin is giving is incorrectly, there is a way you can help with improving the app.
Upload the audio to an eBird list OR onto Macauley Library, and let myself or any other annotator for Merlin know and we can get it added to the training data to improve the system. (Caveat: should be "normal" songs & sounds, no nestlings sounds or obscure rarely heard vocalizations.) We've now got the ability to manually bring recordings into our annotation system, so while this wasn't possible last year it is now!
Upload the audio to an eBird list OR onto Macauley Library, and let myself or any other annotator for Merlin know and we can get it added to the training data to improve the system. (Caveat: should be "normal" songs & sounds, no nestlings sounds or obscure rarely heard vocalizations.) We've now got the ability to manually bring recordings into our annotation system, so while this wasn't possible last year it is now!
