BirdNet-Go Processes RTSP Audio from Three IP Cameras for Local Species Detection

Jason Tucker connected three consumer IP cameras via RTSP to BirdNet-Go, which performed 24/7 inference on-device. The setup detected birds, bats, and frogs while routing alerts through MQTT to Home Assistant and a local Discord channel. Novelty tracking flagged first-time yard species and species-list rules triggered instant notifications.

Data from the deployment shows zero recurring costs and full audio retention on the local network. Perch v2 expanded coverage from BirdNET 2.4's 6,000 species to 14,795. Real-time audio energy visualization exposed microphone placement issues such as wind or HVAC noise, confirming operational reliability on unmodified hardware.

This pattern matches other local inference projects where existing sensors are repurposed for secondary tasks without new silicon. Privacy is preserved because audio never leaves the premises unless the user enables BirdWeather sharing. The fart detection incident illustrates a predictable false-positive class that requires threshold tuning rather than model failure.

Next deployments will likely add multi-model ensembles and Home Assistant entity auto-discovery for occupancy-linked alerts. Integration volume will test whether MQTT discovery scales past single-household installs within twelve months.