I’m a scientist, I like data. If I can collect and analyse data from the environment around me, I’m a happy scientist. I’ve been using BirdNet on my phone to identify birds by their sounds; it works pretty well, but sometimes gets confused by lots of birds together, I’ve taught myself a lot of bird sounds comparing what the app says and what I hear.
I’ve wanted to get an idea for what bird species visit my garden, leaving a phone running 24/7 isn’t really ideal, and anyway the BirdNet doesn’t work with streaming audio, you need to select and submit a period of sound containing the bird you’re interested in identifying. While ponding ways to make it work, I discovered BirdNET-Pi which runs a bird identifier on a Raspberry Pi computer.
I’ve had this running in my garden for two days, it hasn’t really tuned up any major surprises (anything surprising is 99.999% likely to be a misidentification). What it has done is show up the active times of the local birds, the sparrows are a bit over represented as they sit close to the microphone and swamp out fainter birds.
The software claims to have detected 71 species in two days, I’m sure many of these are incorrect, I did have the detection threshold set a bit too low at the start. There are also many identifications of Owls and Bittens that I’ve identified as distant dog barks and wind noise.
At present, the system is a lash up of a bare Pi circuit board and a cheap USB microphone. If I was to leave this running longer I would need to get a proper outdoor, weather proof microphone and a proper case for the Pi. Something I might consider in a few months. It would be very interesting see how the detected species changes though the year.