New Model Branch Testing

[ehpersonal38]

Please put your discussions in regards to the new branch new_model here!

Refer to #184 (comment) for instructions on how to try it out.

[DD4WH]

not sure what went wrong, this is what I get:

I did the following:

set up a fresh base install of BirdNET-Pi on a RPi 4B
connected to birdnetpi.local
adjusted language etc. and other settings
went into putty, connected to my BirdNET-Pi over WiFi and issued the command above
after that, the birdnetpi.local page did not look different and I could not find the setting for the new model

[Svardsten53]

I have now tested with "Species Occurence Frequency Threshold" set to 0.03 and "Minimum Confidence" to 0.5. I played a bunch of species that are here in the winter and I think pretty much all of them were detected. I also played species that are only here in the summer and NONE of these were detected. The post-processing with eBird data seems to work very well also here in Sweden. This looks really good!

[ehpersonal38]

@DD4WH

Thus I propose to add an option to use a custom species list for post-processing == regional post-filtering instead of using the eBird distribution data/model.

This is what Tools -> Custom Species List should do, but I forgot to code it to override the regional post-filtering species list. So I just fixed it, and after a quick test it seems to be working as you hoped, where if the Custom Species List has any entries, the new model will only use that species list for its post processing.

For my understanding, SF_Thresh == 0.01 should result in NO post-processing, i.e. using the whole species available (>3,000 species), but I may be wrong.

So I've been doing some testing with sf_thresh and generating species lists, and it seems that actually 0 means the whole 3000+ species. Contrary to what the BirdNET-Analyzer repo says, sf_thresh can actually go from 0.0005 to 0,99, not 0.01 to 0.99. Which means that it is indeed possible to include rare species on the predicted list while minimizing false positives, or without necessarily needing a custom species list.

One thing I did notice, however, is that the eBird data seems to conflict with the model's data. For example, this week in my area, the White-crowned Sparrow should have a frequency of 0.06 as per eBird:

Yet, the model thinks it basically doesn't ever show up around here, giving it same frequency as some 2,800 other species that aren't in my area:

So, it's not perfect, but it does generally work well for most species. I posit that the sf_thresh can safely be left at around 0.0025 or so (which produces 136% more species than the BirdNET-Analyzer suggested default of 0.05, which is quite a lot.) Definitely needs more testing though.

Also I've pushed a few updates to the branch so anyone on it should go ahead and update for the latest stuffs.

[DD4WH]

@ehpersonal38
Ethan, thanks very much! Excellent work!
Update went well, I will now watch for the behaviour of the model in the next days.
I think we will have to compare a lot between different systems and settings and the two models before we can make suggestions how a user should tweak his/her system . . . :-).

[Svardsten53]

@ehpersonal38
A friend of mine is also testing this in Sweden. He has changed the species list so that each species starts with a capital letter, please include this file in the installation package.

labels_sv.txt

[ehpersonal38]

@Svardsten53 Right, sorry I forgot to do that, will get that implemented ASAP.

[Svardsten53]

Nice work! This is exactly what I was going to suggest - to be able to run both models in the same installation. I will try to test this at my place here in Sweden. I believe that the new model works better here already than what is said, not only in the USA.

I have three species lists - TF-Lite (6363 species), BirdNET-Analyzer (3350 species) and a list of Swedish species (494 species), which I assume mostly matches the European list, but translated into Swedish.

If I match my Swedish list in Excel against TF-Lite, 50 species are missing (10%) and matching against BirdNET-Analyzer results in missing 110 species (22%). In both cases, only a very few species (6 and 17 respectively) are of importance as the others are species that are included in the European list but are not found in Sweden.

I have installed BirdNET-Analyzer in Windows 11 and tested with a self-recorded sound file. If I enter my position in Sweden and week 1 (winter) I get the following message "Species list contains 130 species". If I change to week 25 (summer) I get the following: "Species list contains 146 species". This means that the program takes into account position and time of year and that there are significantly more species than what was said recognized in Europe in BirdNET-Analyzer, but the numbers are still a bit low compared to the findings I made in the Excel search.

[Fish31520]

@ehpersonal38 Hi, I followed you instructions (bash ~/BirdNET-Pi/scripts/update_birdnet.sh -b new_model) and tried to upload the new model but I've got nothing new in "Web interface"/Tools/Settings. There's no "Model" option, nor in the advanced settings.

Any clue ?

[Fish31520]

@ehpersonal38

Here is the output I've got

+ set +x
+ sudo -u Birdnet git -C /home/Birdnet/BirdNET-Pi fetch origin new_model
From https://github.com/mcguirepr89/BirdNET-Pi
 * branch            new_model  -> FETCH_HEAD
+ set +x
+ sudo -u Birdnet git -C /home/Birdnet/BirdNET-Pi switch -C new_model --track origin/new_model
fatal: invalid reference: origin/new_model
+ set +x
+ my_dir=/home/Birdnet/BirdNET-Pi/scripts
+ sudo systemctl stop birdnet_server.service
+ sudo systemctl stop birdnet_recording.service
+ services=(chart_viewer.service spectrogram_viewer.service icecast2.service extraction.service birdnet_recording.service birdnet_log.service)
+ for i in "${services[@]}"
+ sudo systemctl restart chart_viewer.service
+ for i in "${services[@]}"
+ sudo systemctl restart spectrogram_viewer.service
+ for i in "${services[@]}"
+ sudo systemctl restart icecast2.service
+ for i in "${services[@]}"
+ sudo systemctl restart extraction.service
+ for i in "${services[@]}"
+ sudo systemctl restart birdnet_recording.service
+ for i in "${services[@]}"
+ sudo systemctl restart birdnet_log.service
+ sudo systemctl start birdnet_server.service
+ sleep 5
+ for i in {1..5}
+ systemctl is-active --quiet birdnet_server.service
+ grep 5050 /dev/fd/63
++ netstat -tulpn
tcp        0      0 127.0.0.1:5050          0.0.0.0:*               LISTEN      1129844/python3
+ logger '[/usr/local/bin/restart_services.sh] birdnet_server.service is running'
+ break
+ systemctl is-active --quiet birdnet_server.service
+ grep 5050 /dev/fd/63
++ netstat -tulpn
tcp        0      0 127.0.0.1:5050          0.0.0.0:*               LISTEN      1129844/python3
+ status=0
+ ((  status != 0  ))
+ sudo systemctl restart birdnet_analysis.service```

@Datore7 

Thanks, I'll try !

[Fish31520]

@Datore7 Thanks it works very well like that !

[jflyons]

I followed in footsteps of @Fish31520 - the @Datore7 advice got me there! Switched the model in settings and here I go.

[jflyons]

First full day with new model - same system. Notably more detections. Species mix a bit different too. Biggest change at first glance - the European Starlings are better represented!

[jflyons]

A month later - as the birds have become more active my detections reflect it. Also, the ability to detect HOFI, LEGO, and EUST has had step-function improvements, granted I am not ruling out some seasonality effect. Oddly the American Crow has been the most active one to slip under the model's radar much of the time.

[Fish31520]

@ehpersonal38

Strangely, I only managed to switch from the previous model to the new one by changing the Species Occurence Frequency Threshold (with the 0,03 original value, I had a prompt telling me to choose between 0,0205 and 0,0305).

[ehpersonal38]

Thanks for the bug report! I've fixed this, you can update again:

bash ~/BirdNET-Pi/scripts/update_birdnet.sh -b new_model

[Datore7]

Yes, Thanks, works correct now, I did a regular update (the new_model dissapeared), and did reinstall new_model, everything OK.

[Svardsten53]

@ehpersonal38
I would like to know a little more about the species lists and the difference between them. Can you explain why the old list, which is bigger and should contain all the species that the new one has, does not work in the new model? I have accidentally tried to use the old list for the new model and then there were many very strange species. Doesn't the list just work as a translation of the Latin name into the national species name? What happens to the species lists if you switch from the new model to the old one? Does the program take care of this? Thanks in advance.

I am also testing two installations (old and new model) in parallel and will return later with the results.

[ehpersonal38]

Can you explain why the old list, which is bigger and should contain all the species that the new one has, does not work in the new model?

I'm honestly not sure why the BirdNET-lite labels file can't be used for the new model. I thought they could too, and then I tested it, and like you found, there were a bunch of strange species. The analysis completely fails. I don't really know how the model parses labels.txt so I can't say for sure.

What happens to the species lists if you switch from the new model to the old one? Does the program take care of this?

The program does automatically take care of the labels when you switch models. It installs the appropriate labels.txt by extracting the zip file located in BirdNET-Pi/model/.

Hopefully that somewhat answers your questions. Also I'd love to see the results on your 2 installations! Would be great to see the accuracy differences between them.

[Svardsten53]

Thank you very much!

[Fish31520]

@ehpersonal38
Hi,
Yesterday I updated Birdnet-Pi (Tools/System Controls/Update) and since then, the New Model option has vanished from Tools/Settings, and all my detections are messed up (I'm in the south of France and I only get South African birds 😅).
Any idea how to correct that ?
Thanks !

PS : Despite that, the New Model seems to work fine for me... 😄

[Fish31520]

I finally made bash ~/BirdNET-Pi/scripts/update_birdnet.sh -b new_model and a reboot and everything went immediately back to normal, problem solved ! 😅

[ehpersonal38]

Yeah updating via the web interface will put you back on the main branch, but the labels from the new model will not be changed back to the old model, hence the weird behavior.

If you or anyone else ever wants go to back to the main branch:

1. Go to Tools -> Settings, and set the Model back to the old one. Press Update Settings at the bottom of the page and wait
2. Go to Tools -> System Controls, and Update

[Fish31520]

@ehpersonal38 Thanks a lot !

[becausebirds]

@ehpersonal38 This saved me. My model was going crazy and reporting all kinds of random exotic species. Toggling the modules and clicking update resolved the issue and the detections went back to normal.

[Svardsten53]

I have now tested both models during this weekend with identical hardware, with the same settings and with the microphones next to each other. The new model seems to report slightly more detections and more species. The number for each species varies up or down depending on the species, for most the numbers are quite similar. The most positive thing though is that I haven't had a single detection of a species that shouldn't be here at this time of year and that's certainly not the case in the old model. I think this indicates that the connection to eBird also works in Sweden and would think that it also applies to the rest of Europe. I've checked eBird reports for a few different countries and it seems to be about the same level, though not nearly as big as in the US.

I have checked some sounds and most of them seem to be correct. Now I'm not particularly good at sounds, so I've asked some experts to listen and report back with results. I hope to return with this.

The new model works, as expected, even with my adaptations and the translation into Swedish that I made and that most people seem to be using here in Sweden. I suggest that the new model can be implemented in the main branch.

Link to my test system with the new model: https://birdnetpitest.svardsten.se

[ehpersonal38]

This is great @Svardsten53! Thank you so much for testing this out. So it sounds like the model could possibly work pretty well even outside of North America.

It definitely looks like we can get the new model pushed to main soon, I'll just wait a little while longer for other people to test (and I'm still finding and fixing little bugs every day).

[Fish31520]

@ehpersonal38 @Svardsten53
Same thing here in France, the new model seems to work smoothly, and I haven't had misidentification so far...
I don't know if it will work accurately on rare European bird species but for the common ones, it's clearly doing the job.

[Svardsten53]

I just checked my installations and if I use periods these are converted to commas.

[ehpersonal38]

Wow, okay then! I honestly don't understand what's going on there, but I shall let it be. Thanks guys

[Datore7]

Anyway, FYI, when, I look at the log, the positions is shown with dots, i.e. 58.xx 17.xx

[Fish31520]

@ehpersonal38 So far it seems to work well with commas...

[ehpersonal38]

Yeah no worries, I was just a bit baffled it worked with commas! I learn something new every day

[timsterc]

@ehpersonal38 - for the BirdWeather detection posts with the new model, could I ask that that you make one change, so that we're tracking which model is posting detections to BirdWeather :

post_algorithm = ""algorithm": " + ""2p2"" + ","

Thanks!
-Tim

[ehpersonal38]

Sure, that change has been made ~

[Fish31520]

@ehpersonal38

Hi,

I'm wondering if there's a way, in order to check the new model on rarest European birds, to give the new model sounds correctly labeled by the previous model, to check if it gets them right ?

I don't know if such a thing can be done, I'm just asking... 😉

Thanks !

[Fish31520]

@ehpersonal38
I'll let it run for few days at 0,015 and we'll see...
Clearly 0,03 seems to be really tight.

[timsterc]

FWIW that mirrors what we were seeing with the BirdWeather iOS App (which uses the latest BirdNET-Analyzer model). The probability setting frequently needed to be set very low for species that had a higher historical eBird probability for that Lat/Lon/Week.

[ehpersonal38]

@timsterc Thanks for that. Here's an excerpt of an older post of mine:

One thing I did notice, however, is that the eBird data seems to conflict with the model's data. For example, this week in my area, the White-crowned Sparrow should have a frequency of 0.06 as per eBird:

Yet, the model thinks it basically doesn't ever show up around here, giving it same frequency as some 2,800 other species that aren't in my area:

So, it's not perfect, but it does generally work well for most species. I posit that the sf_thresh can safely be left at around 0.0025 or so```

I still have yet to explain why the model's metadata is so different from eBird historical data when given the same exact lat/lon/week of year. Do you have any thoughts?

[timsterc]

I've pointed it out to Stefan - at least for my location - but I'm not sure why the difference. The 'plus' as many of you are seeing is that the high confidence false positives with the new model are fairly rare. On top of the app, I've also been running versions of the new model on my Pi (in Moss Beach, CA) since around March 2022, and it's done a nice job.

[Fish31520]

@ehpersonal38
BTW, how are the frequencies listed with ~/BirdNET-Pi/birdnet/bin/python3 ~/BirdNET-Pi/scripts/species.py --threshold 0.03 calculated ? I'm just curious here ! 😅
They seem to vary from day to day.

[Fish31520]

Hi,

Using the New Model, I recently had 2 different subspecies tagged on the same sample.

Is Birdnet-Pi able to put several label on a 3 seconds sample or is that a small software hiccup ?

Thanks !

[Svardsten53]

As I understand it, several detections and from different species can be processed in the same recording.

[jflyons]

Yes, occasionally I get a "two-in-one" instance like this, any two species.

[Fish31520]

Thanks guys !

[Fish31520]

@ehpersonal38
Hi, very small typo on the new model, but the French name of Chloris chloris is "Verdier d'Europe". 😅

[Fish31520]

@ehpersonal38 Thanks, I'll try that and we'll see... 😉

[Svardsten53]

I have also submitted an issue about the capital letters for Nordic species there, but no reply so far.

[Fish31520]

Strangely, there's always new updates in Tools/System Controls? I don't really understand where they're coming from if nobody's dealing with the troubleshooting ?

[DD4WH]

Have a look here for all the commits of our hard working programmers fixing bugs all night long! :-)
https://github.com/mcguirepr89/BirdNET-Pi/commits/main
Every time there is a new commit, your "Update" button will change colour and prompt you to update . . .

[Fish31520]

@DD4WH
WoW, I just took a quick look at the job done, it's clearly something !
Well, thanks a lot to them, they're doing an incredible job !

[ehpersonal38]

new_model has been officially merged into main! This means that all BirdNET-Pi users should now have access to the new model (if they want to use it) and all the related updates. Thanks to everyone who has helped me test this branch over the past few weeks! Hopefully with all your help we have successfully minimized the amount of bugs introduced. 😄

[Svardsten53]

Thanks, great job!

Is the issue with creating file names when forward slashes are in the common name also fixed?

[ehpersonal38]

Thanks! And yep that's fixed -

[DD4WH]

Great job! Thanks to all programmers involved!
I have tried to put some explanation into the WIKI. It might help in choosing the right model for your needs and adjusting the model parameters according to your requirements.
https://github.com/mcguirepr89/BirdNET-Pi/wiki/BirdNET-Pi:-some-theory-on-classification-&-some-practical-hints

[Fish31520]

@ehpersonal38 @DD4WH
Thanks a lot guys for your amazing job ! 👏

[blogdemoi]

Thank you!

[matthew73210]

Gave the new model a quick wir, installation went fine but it started reporting Crows as Wren-Warblers. And blue tits as grey hooded gulls, which according to ebird have never been sighted where I live. Understandable for the snowy mountains far from the sea. I'll leave it to mature a tad.

[ehpersonal38]

I think the issue is that your labels.txt didn't get installed (this happens after you change the model in Tools -> Settings, then click the Update Settings button at the bottom. You'll see the labels.txt in ~/BirdNET-Pi/model change file size if it works correctly), but yeah the commas in your lat/lon would've prevented either model from working

[ehpersonal38]

ey29yQXLen.mp4

[blogdemoi]

@blogdemoi did you recently update your BirdNET-Pi via Tools -> System Controls -> Update and try out the new model? Are you experiencing the same problems Matthew

@matthew73210
Day 2 after switching to new model. Day 1 was fine and I had about the same detections in my garden, with 1 or 2 false detections. Then at the end of Day 1 I changed the settings to:

-Species Occurence Frequency Threshold [0.0005, 0.99]: 0.03
-Extraction Length: 3
-Overlap: 0.0
-Minimum Confidence: 0.65
-Sigmoid Sensitivity: 1

And I had only 2 detections for Day 2. So I will change the settings back to previous settings (unfortunately I forgot my previous settings :-( what would you recommend for Overlap, Minimum Confidence and Sigmoid Sensitivity? Thks) but will keep new model for next week to see the results.

[ehpersonal38]

Here are my settings:
-Species Occurence Frequency Threshold [0.0005, 0.99]: 0.01
-Recording Length: 15
-Extraction Length: empty
-Overlap: 0.0
-Minimum Confidence: 0.5
-Sigmoid Sensitivity: 1

[matthew73210]

Alls good my end now. Thanks for the help.

[lloydbayley]

Just reporting in from Sydney! The clone, running the new model, of my BirdNetPi is still chooching along just fine. In-fact, it has a curious habit of picking up a few more detections (and they are real, I checked) but I'm now starting to wonder if the new ones microphone is just a smidge better. Am going to do some testing regarding that but there have been no appreciable detection degradation through using the new model in Australia, so it appears to be working for my birds just fine.

[Svardsten53]

Glad to hear it works there too.

I myself now have two identical systems with the new model next to each other, but experience more background noise in one. I have lowered the level in alsamixer and then get slightly fewer detections and species because the sensitivity to low sounds then seems to decrease. I get to troubleshoot sound cards, microphones and more, which shouldn't be a problem when I can swap parts between systems.

[lloydbayley]

Yeh, I have a sneaky feeling there is a subtle difference between the mics I have. They are supposed to be the same but there is a everso-slight shape difference to the USB plug. I think the new one is more sensitive, so I'll do some level tests when I get the change. It sounds clearer in the Live Audio. If it IS a better mic, I'll swap it permanently to the main Rpi and keep the other as a backup. I might even buy another one or two if they have improved them!

[krummrey]

I'm still confused on wether the new model is just for Americans or European users? I'll be recording my 100 000 detection some time tomorrow. A perfect timing to switch to the new model if the consensus is, that it would be an improvement for me here in Germany.
Can anyone clear this up for me?

BirdNET_GLOBAL_3K_V2.2_Model_FP16 (2022)
This model comes from BirdNET-Analyzer, a newer work-in-progress project with aims to improve on the old model. 
Currently it only supports about 3,500 species worldwide, so for users in North America, this model is generally 
much more accurate than the above model, but elsewhere it will be less accurate and possibly useless.

[Fish31520]

@krummrey
Hi, so far the new model seems to work pretty well for me (I'm in France, not that far from you ! 😄).

[Svardsten53]

I live in Sweden and for a few weeks have compared both models side by side with two different RPi computers. The new model clearly provides more detections and more species. Few of the species we have here are missing from the model's species list and there seems to be enough data in eBird even for Europe to avoid false positives.

[krummrey]

Is there a way to train a European model? Take all the species mentioned in Svenson's book and train those?
I have trained image models on Google Colab, but have no idea where to start here.

[Fish31520]

@krummrey
It's seems to be a very good idea, but it's very far from my knowledge ! 😅

[krummrey]

Just a little feedback in numbers: I've switched to the new model on Sunday the 12th.
The three days before I've had 349,555 and 597 detections. An average of 500 per day.
The three days after the switch I've had 1684,2262 and 1495 detections. An average of 1814 per day.
Total detections have tripled while keeping the confidence value at 0.7.

The number of species detected has doubled. While I'm not good at verification I was able to find quite a few false positives.
The carpenter next door working with a hammer was detected as a woodpecker with 92% confidence.
Most of the false positives I found did have a high confidence value. So just raising that won't filter them out without losing a ton of real positives.

Some species do get detected a lot more than others even though they seem to be present at a more equal level visually.

[Vertikar]

Anyone in Australia running the new model and done any comparisons between the two?
Thinking about switching across

[lloydbayley]

I'm running both in tandem but I have a small problem with one of the mics is more sensitive than the other and gain control is not completely accurate. I'm waiting on another mic which will be the same for testing. Normal BNP is running on a Pi4 and the new model is running on a 3B+. I have noticed that the new model picks up more calls than the old one but as I said, it could be the mic. I need to have everything the same for true testing.

Interestingly, there is a dog with a stupid bark up the back somewhere which is erroneously picked up as a Southern Boobook. I don't want to exclude it because I do actually get them around me and I want them recorded, so I just keep and eye on it. The new model has not triggered on that, HOWEVER, I haven't heard/had any real calls so it's hard to tell.
Must chase that mic up. The one I got was DOA so they are sending me another.

My main one is at: https://berowra.birdnetpi.com/ but the other one isn't live, I'm afraid.

[grex023]

Last night two local Tawny owls woke me up in the middle of the night. Out of curiosity I checked if it had been captured but it wasn't showing on the Overview. I checked the logs and I could see it being successfully recognised just not recorded into the database. I'm using the new model with stock settings (Occurence Frequency 0.03) and wondering if I need to change something? This is the second night in a row it has missed it. Lat/Long are set correctly and the owl has been previously recognised on the old model. Any ideas gratefully received

[Datore7]

@blogdemoi
I used this command which I found in this thread: #275

cd && uninstall.sh && rm -drfv BirdNET* && curl -s https://raw.githubusercontent.com/mcguirepr89/BirdNET-Pi/main/newinstaller.sh | bash
It worked for me, took a while, had to enter postition and password, I think it is descirbed in the thread..

But most important, all data in BirdSongs is untouched.

[ehpersonal38]

I'm sorry about this - I committed some code to the repo which seems to have broken things. I've reverted the change so an update should bring your installations back to a working state.

[Datore7]

Thank you @ehpersonal38 !

These things happen sometimes, but no danger on the roof.
I did my update of 2087 updates :) , went fast and well.

[blogdemoi]

It doesn't matter for me, I learned how to boot from a USB key instead of a SD card :-)

[blogdemoi]

Thank you Datore7

[srchild]

Seems the same problem as my thread #740

My apprise code seems to be an incorrect version. It does not have proper initialisation of plugin_paths for apprise

I wonder whether the 2081 updates were some sort of regression to an older version / older codebase. I was thinking of updating apprise manually but wondering whether there is something more than just that which has gone wrong.

[srchild]

Hmm. So I did a reinstall

cd && uninstall.sh && rm -drfv BirdNET* && curl -s https://raw.githubusercontent.com/mcguirepr89/BirdNET-Pi/main/newinstaller.sh | bash

Now my installation is running again. BUT!!!

The reinstall wiped BirdsDB.txt

Fortunately I had made a backup. I've copied the backup over the empty BirdsDB.txt, and rebooted, but my installation frontend is not showing any past detections

BirdDB.txt is there, as are all the birdSongs. How do I get them to be detected and displayed? Or is it something that will rollover at midnight?

[srchild]

Detections from earlier this morning are showing, so presumably the lack of display of past data might be remedied at midnight?

[drinkdhmo]

The location information seems to have an issue. I've put in a location in the Northeast US:
Latitude: 41.0000
Longitude: -77.5500

And the species list tester returns a bunch of birds in South America:

Getting species list for 41.0000/-77.5500, Week 16...INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
loaded META model
Lurocalis semitorquatus_Short-tailed Nighthawk - 0.96928275
Asio otus_Long-eared Owl - 0.9118327
Cyrtonyx montezumae_Montezuma Quail - 0.9020251
Acrocephalus paludicola_Aquatic Warbler - 0.88012236
Cacomantis sonneratii_Banded Bay Cuckoo - 0.8683705
Megaxenops parnaguae_Great Xenops - 0.8516968
Campylopterus falcatus_Lazuline Sabrewing - 0.80343974
Anthus trivialis_Tree Pipit - 0.7810185
Hemitriccus inornatus_Pelzeln's Tody-Tyrant - 0.7620898
Lamprotornis superbus_Superb Starling - 0.7247968
Inezia inornata_Plain Tyrannulet - 0.70437914
Cygnus melancoryphus_Black-necked Swan - 0.6513784
Lanius vittatus_Bay-backed Shrike - 0.64458853
Anthracoceros malayanus_Black Hornbill - 0.644369
Iole propinqua_Gray-eyed Bulbul - 0.63639456
Centropus chlororhynchos_Green-billed Coucal - 0.6204142
Alaudala raytal_Sand Lark - 0.5522935
Hydropsalis cayennensis_White-tailed Nightjar - 0.52345216
Icterus galbula_Baltimore Oriole - 0.5184016
...

[ehpersonal38]

I can't replicate this on my system by setting my lat/lon to yours and generating a species list with the tester tool.

I recommend making sure your system is on the latest update (System Controls -> Update), then "refreshing" your model by changing it to the old one in Tools -> Settings -> Model, saving your settings, then finally changing the model back to the new one. This will hopefully reset your labels.txt to be correct.