BirdNet-Go turns home security cameras into a 24/7 bird identifier

Jason Tucker, an IT professional in Los Angeles who writes about his homelab, wanted to go one step beyond the phone app his wife uses to identify birds. The result is a system that listens to his garden around the clock: three of the security cameras already mounted around his house now feed their microphones into BirdNet-Go, an open-source soundscape analyser, and identify birds by their song.

From security cameras to a 24/7 bird identifier

The software runs in Docker and needs no special hardware. Most modern IP cameras expose an RTSP stream, so the existing cameras simply point at the analyser; Tucker says the container was up and detecting in about fifteen minutes. Because it listens constantly, the system logs an identification the moment a bird starts singing, with no manual recording. It also picks up bats around dusk and, to his surprise, frogs.

Local AI models, with no cloud and no subscription

All inference happens on his own hardware — a server or a Raspberry Pi — with no cloud services and no API calls, which keeps the audio in the house and removes monthly costs. BirdNet-Go supports multiple models, so different detection strategies or regional databases can run side by side. The model gallery recently added Google's Perch v2, raising coverage from roughly 6,000 species with BirdNET 2.4 to 14,795 species. As a privacy measure, the microphones cut out when they hear speech.

Alerts, novelty tracking and shared observations

Detection rules can match against species lists: Tucker gets a notification when a bird he cares about appears, and the alerts land in a dedicated channel on his home Discord server. The novelty feature flags species that are new to the yard and keeps a running list, which he describes as turning into a game for the household. Optional BirdWeather integration shares detections with a community dataset, and Home Assistant can discover the system over MQTT. Third-party projects extend it further, including Saezuri, a self-hosted display that arranges recent species into a woodblock-style collage, and BirdNET-Go Companion, a free iOS app by Robert Oesterlin.

A year of detections — and one odd notification

Over the last twelve months the setup recorded 418,726 detections covering 271 unique species, at an average confidence of 60.9 percent. The most frequent visitor in his part of Southern California is the House Finch, with 118,667 detections. Not everything was a bird: one night Home Assistant reported an unexpected sound in the driveway after a neighbour walked past. Tucker's conclusion is that the project outlasted his expectations — the cameras he bought for security now answer a question he and his wife check daily.