Field guides in the future (1 Viewer)

[jurek]

I'd love a shazam app for bird calls. I mean you can have all the calls of every bird in say, Ecuador, and then u hear an unfamiliar call you'll not even know where to start looking, let alone play it back!! This is why local guides are worth the money

You have no idea how many people tried this over several decades and failed. Last attempt from Cornell works more in 50% of cases if you have an easy call really close. Poorer quality recordings, contact calls and complex songs fail hopelessly. The same with bird ID.

Maybe more sensible would be a reverse, an app constantly listening and throwing possible bird names matching the sounds around. Especially if it would be able to replay the sound which tempted it, and replay the proper bird call for you to match one with another.

 

[pppppvvvv]

Certainly, in the bat sound analysis world, there is auto i.d. software, becoming quite widespread. And the halfway point of the software giving you the call parameters, letting you decide the species, is equally useful.

But with bats, we have come from peak frequency of X, call duration y, call interval z, max and Min frequencies etc. All easily grabbed by software..

For birds, we have, "a harsh nasal tschurrr"! Or " a dripping tap".

There is probably a PhD waiting for someone to treat bird calls like we process bats.

 

[Deb Burhinus]

Or, something like Google's voice commands, where the app just listens, and tells you what it's just heard... "Chizik"... "Pied Wagtail", that would be cool

I was actually thinking going a few steps further

A.I. combined with optics - google gets together with Swaro to produce ‘google bins’ that have in built AI functions that have a voice activated display, not only identifying the bird you are pointing your bins at on command but also having a ‘save’ option to upload to your preferred platform/App. Of course, it would have a ‘what am I listening to?’ as well as a ‘what am I looking at?’ function. An ideal optic for those who already rely on bird call apps and apps like Merlin to ID birds.

 

[njlarsen]

I assume you are aware of this gadget which is half way there: https://www.swarovskioptik.com/us/en/birding/products/binoculars/dg

Niels

 

[Deb Burhinus]

I assume you are aware of this gadget which is half way there: https://www.swarovskioptik.com/us/en/birding/products/binoculars/dg

Niels

No, I wasn’t at all aware of that Niels 😯. But let’s face it, gadgets are really mostly a ‘guy thing’ I think so I rarely pay attention to anything other than my one pair of bins and scope as aids to field experience!

This is amusing though in the blurb on the link:

“The dG helps you to identify birds and other animals on your own”

It should read: “The dG identifies birds and animals for you when you can’t be bothered to work them out for yourself” 🤪

 

[jurek]

Certainly, in the bat sound analysis world, there is auto i.d. software, becoming quite widespread. And the halfway point of the software giving you the call parameters, letting you decide the species, is equally useful.

Luckily for bat people, there are about 10 times fewer bats than birds, and their main call during echolocation is a relatively simple sonar.

I think with birds, one needs a good programmer to code actual features which birders use for ID, for example 'last but one sound should be high, and the last low' in Chaffinch song, and make training sets of difficult pairs or calls. I suspect bird sound ID is instead done by a standard method, basically somebody throws a sound library into software, and hopes the software figures everything by itself. Which it does not.

Ditto with bird ID by sight. I am not sure a software can even work like a birder - first parse a picture into bird head, bill, wing coverts, primaries, secondaries etc., and check for key differences.

Another cool thing would be a software listening for background sounds. Humans can pick a bird sound far away by picking few loudest sounds at the edge of hearing and spaces in between. Which is an enormous help in finding uncommon birds. Software could do it, too.

 

[njlarsen]

It should read: “The dG identifies birds and animals for you when you can’t be bothered to work them out for yourself” 🤪

This might actually come from the developers of Merlin who are aware it has limitations
Niels

 

[Andy Adcock]

Certainly, in the bat sound analysis world, there is auto i.d. software, becoming quite widespread. And the halfway point of the software giving you the call parameters, letting you decide the species, is equally useful.

But with bats, we have come from peak frequency of X, call duration y, call interval z, max and Min frequencies etc. All easily grabbed by software..

For birds, we have, "a harsh nasal tschurrr"! Or " a dripping tap".

There is probably a PhD waiting for someone to treat bird calls like we process bats.

I made this comment in relation to some of the Swiftlets which are practically, impossible to ID in the field and end up being ticked by most, on range alone. Slightly more applicable to Swiftlets though as they do echo locate like bats do.

 

[dantheman]

Ditto with bird ID by sight. I am not sure a software can even work like a birder - first parse a picture into bird head, bill, wing coverts, primaries, secondaries etc., and check for key differences.

With advances in AI and the technology being there (exponential curve) no reason why it won't be long before recognition based on a suite of id pointers and features as is done in humans can't be done, if not quite at the same level at a reasonable expense.

Probably could be done already if someone (Nasa?) threw the resources at it.

 

[pppppvvvv]

In fact, I've just ran two calls thru bat detector software. (I had to pretend the calls where in the typical bat range, so the frequencies and lengths are all 10x out.)

Top one, with mouse pointer is willow warbler call, bottom one is chiff chaff call.

That software would have no issues separating them. Songs would be potentially be easier, as there are typically more calls in the sequence.

Andy/anyone: if you want to give me some examples of near identical calls between different species, I'll look them up on xeno canto and see what comes out. Preferably Calls as opposed to full on bird song!

For the examples below:
Ignore the chiff chaff call distance of 0 ( it was a single call)
We all know the call length will be longer for willow warbler, and a different shape.
But the higher max Freq., Min Freq., And peak freq? My ears/brain wouldn't comprehend that.

The recordings don't have to be particularly good for accurate data to come out. Plus I'd bet in many species, the calls are fairly consistent between individuals.

 

[YuShan]

Using AI to identify birds based on songs, calls or appearance would for me take the fun out of birding. I don't even like to use guides. The fun is in finding and identifying the birds by myself, rather than let somebody else do the hard work.

 

[opisska]

I'd love a shazam app for bird calls. I mean you can have all the calls of every bird in say, Ecuador, and then u hear an unfamiliar call you'll not even know where to start looking, let alone play it back!! This is why local guides are worth the money

This already exists - BirdNet from Cornell is one example. It's free but needs online connection. I have only tried it in Poland and it works pretty well, not sure if it would handle tropical habitats with the huge gamut of possible species.

 

[Swissboy]

Using AI to identify birds based on songs, calls or appearance would for me take the fun out of birding. I don't even like to use guides. The fun is in finding and identifying the birds by myself, rather than let somebody else do the hard work.

That's what is so great about birding. We can all do it the way we like it, taking also into account what we can do well or where we are struggling (like me having a terrible sound memory).

 

[pppppvvvv]

Using AI to identify birds based on songs, calls or appearance would for me take the fun out of birding. I don't even like to use guides. The fun is in finding and identifying the birds by myself, rather than let somebody else do the hard work.

I agree, I try to stop short of the full AI, but collect the data, and try and work out the species myself. For bat calls at least.

But if two near identical species can be identified in the field by playing their call into an app which gives you the max frequency, and enables you to eliminate one, that would be cool.

 

[pppppvvvv]

This already exists - BirdNet from Cornell is one example. It's free but needs online connection. I have only tried it in Poland and it works pretty well, not sure if it would handle tropical habitats with the huge gamut of possible species.

Wow, that is really good. All pretty much live as you hear the bird. No downloading to a pc, and sifting thru files when you get home!

For me, this app would be the best way to learn calls too. Do we know how extensive the library of calls is? Edit: currently, 894 common species of North America and Europe

 

[jurek]

With advances in AI and the technology being there (exponential curve) no reason why it won't be long before recognition based on a suite of id pointers and features as is done in humans can't be done, if not quite at the same level at a reasonable expense.

Probably could be done already if someone (Nasa?) threw the resources at it.

It can already be done in software, simply needs time.

It would need to be a labour of love of some highly paid data scientist who is also a birdwatcher. Or a clever splitting the job between birdwatchers marking the important features of bird songs, and a programmer coding them. I assume nobody did it and opted for an easy approach (lets run a program on everything).

Andy/anyone: if you want to give me some examples of near identical calls between different species, I'll look them up on xeno canto and see what comes out. Preferably Calls as opposed to full on bird song!

Try short high calls of great, blue and coal tits and goldcrests? Or alarm calls of small songbirds?

This already exists - BirdNet from Cornell is one example. It's free but needs online connection. I have only tried it in Poland and it works pretty well, not sure if it would handle tropical habitats with the huge gamut of possible species.

I tried it playing xeno-canto sounds from my computer and it failed with complex songs, uncommon species, and weaker recording. Too bad to be useful in the field, even forgetting that one cannot rely on internet connection outdoors. But I guess it could be near the top of what can be achieved with a limitation of not giving song features to train the program, and a training set of ca 10000 of sounds for several 1000s of birds.

Such programs also usually give completely stupid answers to sounds from outside the library, for example a non-bird sound or a rarity. It is because most algorithms develop a number of fixed categories and try to match to one of those.

 

[lmans66]

Is a 'field guide' a pocket book that you can take into the field, or....more a bible that sits on the shelf?
Or ...is your thinking of a field guide more akin to being on the Internet, making it more a 'birding resource' than a field guide? jim

 

[pppppvvvv]

I think the thread, originally, was discussing how field guides might become more dynamic in the future, with apps, and online resources, mobile phones, internet etc.

I suspect we have pushed out on a tangent with talk of AI etc. But you know what, this is one of the most enjoyable online discussions I've had for ages, so keep it going! For the sake of sanity!

 

[dantheman]

It can already be done in software, simply needs time.

It would need to be a labour of love of some highly paid data scientist who is also a birdwatcher. Or a clever splitting the job between birdwatchers marking the important features of bird songs, and a programmer coding them. I assume nobody did it and opted for an easy approach (lets run a program on everything).

You realised I was talking about visual? Like the Merlin App etc in real time. Could probably zoom in and id things at further range than a human ... knowledge, probabilities, jizz etc

 

[dantheman]

I think the thread, originally, was discussing how field guides might become more dynamic in the future, with apps, and online resources, mobile phones, internet etc.

I suspect we have pushed out on a tangent with talk of AI etc. But you know what, this is one of the most enjoyable online discussions I've had for ages, so keep it going! For the sake of sanity!

BirdForum always aims to please ... it's gotta make up for all the real life threads from the fun year that was 2020 ... ;-)

(Give it 20* years and you won't know if the keen young birder out finding stuff on your patch is actually a human or just a humanoid ... or is that getting depressing again ?!? - do robot birdwatchers dream of electric cattle/intermediate egrets? )


*(ok maybe not, but only 2 or 3 if they are only on social media??)