BirdBuddy: an offline bird call identifier that runs entirely on your device

Description and use

BirdBuddy is an offline bird call identifier that runs entirely on the device, using BirdNET for acoustic classification, Piper for TTS, and a fine‑tuned Qwen3.5‑4B for expert notes.

Problem

Hikers and birders need to identify bird calls in remote locations without internet access.

Outcome

BirdBuddy delivers offline bird call identification with a 52 MB classifier and 63 MB TTS voice, achieving <2 s inference latency on a laptop CPU and no external dependencies.

Workflow

• Open the app
• Tap once to record a 3‑second bird call
• The app runs BirdNET ONNX inference locally
• Piper TTS speaks the top match
• Optional: adjust Top‑K, confidence threshold, or voice speed in Settings

Evidence

BirdBuddy tells you what it is — with everything running locally on the device in your pocket.

Source

The identification finishes in about two seconds. No cloud round‑trip, no account, no upload.

Source

I used the FP32 truncated‑DFT ONNX variant (~52 MB) running through ONNX Runtime on CPU.

Source

Piper speaks it out loud: "I heard a Northern Cardinal."

Source

I built a 165‑example chat dataset covering 60 common North American species, then fine‑tuned Qwen3.5‑4B with a LoRA adapter (r=16) using Tinker.

Source

Copilot was the workbench for the entire project

Source