Or you could use my sbts-aru project. Which is free software that runs on Raspberry Pis. It also allows you to perform sound localization remotely if required, without bringing the SD card back. Last night I had a listen to some calls in the jungle from Lajuma research center in South Africa that was recorded with them. @nlubcker  and I expect to do some localization tests soon from the species down there.



If your Pi is connected to a network you can do localization in near real time (less than 10 minutes manually) as you don't have to stop the recorder or post process the recordings. If you write you own pipeline you can implement real-time sound localization with it and output a URL to a google maps location.



 

Read "GitHub - hcfman/sbts-aru: Low cost Raspberry Pi sound localizing portable Autonomous Recording Unit (ARU)" on GitHub

GitHub - hcfman/sbts-aru: Low cost Raspberry Pi sound localizing portable Autonomous Recording Unit (ARU)

Low cost Raspberry Pi sound localizing portable Autonomous Recording Unit (ARU) - hcfman/sbts-aru

GitHub

My advice for microphones is the em272 microphone capsules based ones, which are very high quality microphones used as by the Swedish company Telinga in their parabolic microphones. That's what I use. They are very low noise and very sensitive. Here's a link to one, likely hard to get at your side of the world though.



 

Read "Clippy EM272Z1 Mono Microphone - Veldshop" on Veldshop

Clippy EM272Z1 Mono Microphone - Veldshop

A clip-on lavalier microphone using the superb low noise, high sensitivity Primo EM272Z1 omni capsule.

Veldshop

In my testing they appeared to be similar in performance to some Rode clip microphones and similar in price. They are likely more easily obtained where you are.

Como dijo Carly, Audiomoths y Song Meter Micro, son los que mas te recomiendo. lo unico a tener en cuenta es que para audiomoths vas a tener que comprar o armar alguna especie de cajita protectora. los de Wildlife acoustic, ya viene con protección, solo recomiendo hacer un "techo" con algun plastico. 

This is really impressive Avi! Would you mind sharing what kind of hardware the local ingestion + detection is running on? 

It's a fun experiment to think if this could be expanded to any live stream happening on any platform (Instagram, TikTok)

What about this year? Who will be there? 

https://www.naturetechweek.com/

I am planning to be there for the Unconference and some satellite events.

Brett, you may reach out to @Lucille, who under the auspices of the Partnership  for Observation of the Global Ocean (POGO), the Scientific Committee of Ocean Research (SCOR), and the International Quiet Ocean Experiment (IQOE) is managing the development of LC-MARE a low cost marine acoustic recorder.  An ultra low power ADC is indeed an important component, so there may be some synergies.

This is really interesting — especially the part about experimenting with different high-value uses.

I’m currently working on a small edge AI project for ecosystem monitoring, and it’s made me realize how different things can look outside of controlled environments.

Out of curiosity, during those experiments, what ended up being the biggest constraint — was it more about technical feasibility, cost, or something unexpected in real-world conditions?

cool! 
When you say: "...first usable result that validates our hardware and software signal chain." can you share what those chains are? 

As for connectivity, yeah, that's a HUGE challenge ... Can you do SMS-level connectivity to Starlinks?

I presume you'd want to log basic temp/salinity, and perhaps include a basic accelerometer for wave motion over something like MQTT. 

As for pushing down bioacoustic processing to the devices, yeah, would be awesome to do some minimal envelope threshold detection, and just send back compressed versions of 'the good stuff' .  Maybe even listen for propellers and/or fishing sonar too, and send EarthRanger alerts.

And of course you'd want to do OTA updates, which in turn introduces security, etc.

Fun project!

Hi I am also looking for collaborators

Read "Looking for Collaborators: AI-Powered Backyard Bee Monitor for Citizen Science | WILDLABS" on

Looking for Collaborators: AI-Powered Backyard Bee Monitor for Citizen Science | WILDLABS

if interested in the project let me know

we can add acoustic monotoring for bees

Hi Leonardo,

Let’s chat! I potentially have some opportunities, collaborations or even short-term employment. Send me a message. 

Hey Leonardo

I might have something. Lets chat, reach out to me at [email protected]

I recently was at the Los Amigos station, and just for fun turned on my Audiomoth at high frequency (~100khz) and got some crazy frog chorus recordings (along with bats and insects, etc) at a mosquito-infested pond in the Amazon. I too am indeed am looking for where to begin...

Hi Stephen, you should connect with the acoustic team at WildTrax (environmental sensor data management platform used across Canada)! They can definitely answer both of those questions. You could search the Data Discover portal (on WildTrax - https://portal.wildtrax.ca/discover) to see who has ARU recordings that might interest you. There is lots of open data on WildTrax.

You can also use tools like HawkEars to scan ARU recordings for hundreds of Canadian species, including amphibians (HawkEars is embedded in WildTrax, too): https://github.com/jhuus/HawkEars

Hope this is helpful!

Hi Fatin,

this sounds very interesting. I am an AI engineer and I just finished a job where i developed an AI model for audio classification and deployed it on an AudioMoth. Please let me know if you are interested in any kind of help or collaboration!

Leonardo

Greetings Nurfatin, it is interesting to see to the project you are working on.  If there's anything I can do to help with my skillset let me know.

Thanks, Mike

Hi, I am trying to identify caimans' sounds. A friend of mine is trying to identify primates' sounds in South America.

Hi Danielle,

I haven’t trained BirdNET directly, but I recently worked on a multi-species bird classification project using custom labeled datasets.

In our workflow we trained lightweight models using data augmentation and knowledge distillation from larger models, with the goal of making the system suitable for deployment on low-power devices like AudioMoth.

I’d be very interested to hear about others’ experiences adapting BirdNET to new taxa, especially bats or insects.

Just now seeing this - so apologies for the late reply! We studied multiple different bioacoustics embedding models (including BirdNet v2.3) for transfer learning performance for underwater and marine mammal tasks. The BirdNet version we examined had solid performance alongside Perch 1.0 and 2.0 models on the tasks we evaluated  (see: paper - Perch 2.0 transfers 'whale' to underwater tasks and blog post for more details). For example code - we have this end-to-end colab demo for an agile modeling workflow to create custom classifiers where there are multiple embedding model options to choose from (including BirdNet). Hope this helps!