Beginner questions: static on RTSP feed and New Zealand bats?

[adamshand]

Hello,

Thank you so much for your work on birdnet-go, really cool project! I've got my first Raspberry Pi up and running with an AudioMoth microphone.

A couple beginner questions:

• Is it normal for the sound files to have a lot of static? Like listening to an untuned radio station with bird calls over the top (attachment below)? Or does that mean something is wrong? Can reduce the static by doing some filtering in Settings > Processing, but is that expected? It seems to be detecting birds just fine.

• Interested in trying bat detection, but I'm in New Zealand and there aren't any NZ models list yet. Is it worth trying one of the US etc models? or is it only going to work with a model trained for our species?

Thanks!
Adam.

anthornis_melanura_27p_20260608T070723Z.mp3

[birdnet-go-bot]

Automated message from birdnet-go support assistant.

Yes, a lot of static or hiss is normal with the AudioMoth. It is a highly sensitive microphone designed to pick up faint sounds, so the noise floor is quite noticeable when you listen to the raw audio. If it is detecting birds fine, I recommend leaving the filtering alone. Heavy noise reduction makes it sound better to human ears but can actually degrade the signal for the AI.

For bat detection, you can try one of the other models just to see if it picks up activity, but the species labels will be wrong. BattyBirdNET models are trained on specific regional species. Since there is no New Zealand model yet, a US or EU model might trigger on a New Zealand bat because the echolocation clicks sound similar, but it will confidently misidentify it as a foreign species.

[adamshand]

Perfect, thank you!

[stevedee]

In NZ there are just 2 bat species, both are unique to NZ.

Whether its worth using bat detection depends upon your interest; if you just like collecting species lists, definitely not. But if you are interested in bats, definitely yes,

Bat calls are very distinctive, even if they may appear similar from one species to another.

Echolocation consists of a repeating train of calls, occurring in a band of frequencies within the ultrasonic range.

When they spot a potential food source, the calls become closer & closer together as they zoom in. Then the frequency & power of their call drops during the final milliseconds of their hunt. This is the classic 'feeding buzz'.

They also seem to find time to chat to one another. The 2 very loud 'social calls' in this trace, overload my system. Its a good job we can't hear in this frequency range, as it would be bloody deafening!

If quality, reliable calls are available for NZ species, you may be able to find someone willing to produce a species list for you. But start with any list & see what happens.

Looks like an interesting journey ahead!

[adamshand]

Great information, thanks! Will give it a try with one of the models and see if we detect any bats!

[zandr]

I'm across the ditch in Sydney, and would love to figure out how to build a model for the bats we have here. https://ecoacoustics.tern.org.au/ has a large dataset for AU, and a few hundred samples for NZ, though only 17 identified for one of your two species.

I'd love to figure out how to build a new model for BattyBirdNET, but haven't put the time in.

[adamshand]

Ooh cool. I’ve never trained a model, it can’t be that hard right? If you have a go, report back please!

[tphakala]

I have written a guide for classifier training at https://github.com/tphakala/birdnet-go/wiki/Training-a-Custom-Classifier, however I don't know if process for BattyBirdNET is different. Maybe @rdz-oss could help?

[rindlerblabla]

I am training my own model (for Swedish species/noise for my setup/microphone) and I think the procedure is the same. However as I have understood you have to convert the tflite-file to onnx with https://github.com/tphakala/birdnet-onnx-converter and then replace a bat model which is already installed, if you want to keep perch and birdnet for birds-detection.

[stevedee]

I hope you do explore the NZ bat-scape with your detector, even without auto species id. Its possible that you may be able to tell the 2 species apart (if both are active in your area) by examination of the spectrograms.
Bats of different sizes beat (flap) their wings at different rates. The interval between echolocation 'shouts' is generally related to wing beats; they shout on the down-flap (rather like the grunt of some tennis players when they serve). So (if you are lucky) the average interval may differ between the 2 species.
Also the shape of the call may differ between the 2 species; l or \ or n...
I have a lookup sheet built into my bat detector version, but this is of no direct use to anyone outside the county of Sussex in England. But just for illustration, it looks like this...