Hi folks. I'm not a serious birder by any means. Last week I joined up to ask what a Spotless Starling was as my Merlin app identified its call. WRONG! Seems it was a misidentification of a common Starling doing impressions. Well this morning my easily fooled App heard an American Pipit and a Short Toed Treecreeper! In Formby! I can't wait for it to hear a Kiwi. I don't think I'll give it my full trust any more!
-
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.
Merlin misled? (1 Viewer)
- Thread starter Tapes
- Start date
PYRTLE
Old Berkshire Boy
Ditch it and study a UK field guide together with some sound and calls recordings. And far more satisfying to assess for yourself ( in most cases ).
Andy Adcock
Worst person on Birdforum
Is there no regional setting which could eliminate these errors?
That does not jibe with my experience: you have to choose which bird pack you have as active in Merlin. I guess you had used it in India and had forgotten to change the bird pack back?
You should not give it your full trust, but I have frequently been able to find no 1 or 2 on the list of likely candidates after a Merlin use.
Niels
Oldnintheway
Well-known member
I gave Merlin a try last year while in Costa Rica and it failed miserably, with fairly common birds. I won't bother again.
For what it's worth.. I have actually good experience with it here in Poland.. I have used it quite often for the past half year or so, just to have it running in the background while birding in a stationary place.. So far I haven't experience any exotic birds and most of the times it get the species right.. Mostly common birds like blackbird, sparrows, tits etc.. So far I only noticed that it constantly recognize a Crested Tit as a Snow bunting for some reason..
I have installed the Europe: Western as a standard package as the only one.. Maybe that helps to avoid exotic species..
I have installed the Europe: Western as a standard package as the only one.. Maybe that helps to avoid exotic species..
jmepler
It's just a flesh wound.
This is the biggest issue when users don't allow Merlin to access their location. With this access, Merlin is able to search through a much smaller list of potential options, so you are much more likely to get the correct answer. Or, at least a more plausible answer. This is a work-in-progress, and getting better all the time.
LittleBitOfBreadNoCheese
Well-known member
As discussed before, Merlin Sound ID doesn't use bird packs, it uses a standalone neural network with around 1000 species.
I have been using extensively Merlin since a couple of years, perhaps three, specially after a large pack of European species was added to the list of species supported, boosting its performance in my area. I wish I would have had such a wonderful aid for learning bird sounds on spot when I started birding long ago, so I am very grateful to Cornell! My hearing is far from good, and many times it alerts me about the presence of a particular species allowing me to detect it visually, or by hearing if they get closer/louder. As with every tool, common sense in using the app is important. I do not pretend, for example, to rely on a tentative ID given on the pair Common/Spotless starling here in winter, or to get every single "chirp" IDed correctly. In some situations I can "hear" and ID some species directly from reading the sonogram even if Merlin is not ID-ing them (i.e. firecrest). In other cases, playing the recording close to my ears confirms the species I haven't been able to hear "live". Understanding how it works is also important to know its strenghts and limitations. Short calls are difficult, mimicry difficults correct ID, while in a chorus with several species the app usually is in advantage. For what it is worth, I use it in a daily basis at Spain with the "Spain" package installed on Ebird and always using positioning (tracking) in my device, and cannot recall to have have been suggested an American species (after the mentioned improvement, before I got them), although sometimes makes obvious false calls. It also still lacks some important species here (it is useful to have a look at the supported species list from time to time). But in the whole is fairly consistent. I would never expect a 100% success (as I don't also with many guides, etc...) Of course, each one has its own opinion and birding circumstances, abilities and limitations. The above ones are my shoes...
Last edited:
LittleBitOfBreadNoCheese
Well-known member
Stefan Kahl who wrote the birdNET AI mentions Starling specifically as a challenge in his paper:
"The poor results for species known to incorporate hetero-specific material into their vocalizations like the European starling (Sturnus vulgaris) and northern mockingbird (Mimus polyglottos) implied that imitation is a significant challenge for automated recognition systems and scores remain low despite good quality training data. "
Mind you he submitted this paper in late 2020 so it maybe that he and the Merlin team have made progress on it.
"The poor results for species known to incorporate hetero-specific material into their vocalizations like the European starling (Sturnus vulgaris) and northern mockingbird (Mimus polyglottos) implied that imitation is a significant challenge for automated recognition systems and scores remain low despite good quality training data. "
Mind you he submitted this paper in late 2020 so it maybe that he and the Merlin team have made progress on it.
