BirdNET Pi - Birding Bot

Stephen T Bird

My excitement of a much appreciated, recent care package seemed worth adding to the BirdNET-become-Merlin sound ID discussion.

The first results are attached as an image.

BirdNET Pi, what is it? 
In short, a simple way for anyone with an internet connection and a little effort to have an automated bird listening station, relatively cheap.

In detail, it’s a RasperryPi (programmable micro controller) with a lite version of BirdNET loaded onto it, software known as BirdNET Pi, , with a microphone attached. The setup is fairly easy. You buy a Raspberry Pi (model 4B, 400 (same as a 4B but inside a keyboard), 3B+, or OW2), connect it to the internet via an Ethernet cable (or WiFi with slightly more effort in command line) and a monitor, enter a single line of code to install BirdNET Pi, attach a microphone and away it goes! You visit the local webpage through your web browser (attached image) on any device that’s connected to your local network.

Photo attached of the results of the first few hours (note the count is observed intervals, not individual counts).

Sure it still runs the same errors of BirdNET/Merlin. RaspberryPis too are running into the same chip shortages and global supply chain issues (you can still get 400s). Due to those chip limitations and my limitations of time, the friend who sent mine up was kind enough to pre-load mine from a Pi he had, and one of his San Diego tests yielded a dog identified as an Elegant Trogon* (not the worst judgment, going back and listening to the dog at a distance…). And it’s still missed a dozen or so common species I’ve heard farther away out the window where it’s sitting. I still need to work on my microphone placement (currently in a medium sized lemon tree in a Berkeley/Oakland  backyard near a water feature).

But! what an excitement it’s been to see the rhythm of the day, as the groups pass by, checking out the water, including the gnatcatcher that infrequently stops by. You can go back and listen to each identification, and it only counts high confidence ratings (*see above Elegant Trogon). What a change from even just 10 years ago and the same friend and I were coding our own neural networks to achieve a lesser result! It’s close to plug and play. I imagine the possibilities of teaching/learning (going back to verify results) and the possibilities of geography and high resolution data (whether across a state, county, or a backyard): democratizing technology.

for others uploading to the web

Best, and I’d love to hear others experience,