May 24 2022
New Project: Birdnetpi
I’ve been messing with a new project to listen to the bird calls in the yard and identify them using Machine Learning. I found this project:
And I built it out:
The microphone choice was painful! The advice on the site is to get a Clippy – a UK manufactured omni-directional mic. After ordering that, then a comedy of adapters, I found that the UGreen USB Audio adapter that I got would not show up as a recording device on the Raspberry Pi. So I punted, googled around for similar projects, and saw people liking this Samson USB mic that has adjustment for close or omni sound.
https://www.amazon.com/dp/B001R76D42?ref=ppx_yo2ov_dt_b_product_details&th=1
And it works! Here’s one day’s data:
I didn’t fire it up til right before work at 8a, so I’ll see if there are more morning captures tomorrow. Interesting how it all goes quiet when the sun goes down.
It’s fun to watch the running log. The software does 15 second sound captures, then runs that sample through the embedded BirdNet machine learning model that contains over 3000 songs. It calculates the confidence level, and anything over .7 gets added to the database.
We knew the crows were loud and seemed to show up at certain times, now it’s documented! 😉
Things to mess with now:
Weatherproof the setup, but don’t block the mic.
Try other placements in the yard for more songs, maybe set up bird feeder closeby.
Understand how the machine learning model was built and is used.
Add other sounds to the model, especially the jet sounds as they pass – perhaps identifying type of jet?
See if the Clippy would be a better mic. Have to figure out the mess of adapters I got to make it connect. Also have to get a proper USB Audio recording device – UGreen looked right from the Birdnet site, but I may have gotten the wrong one.
UPDATE: added a full day chart:
