Record audio from your detector coil, label it, and train a model that learns your specific detector's signals for coins, trash, and ambient noise.
Run real-time audio classification through a CNN model as you swing the coil. TARGET, JUNK, and AMBIENT classifications update live.
Capture labelled audio samples with depth, soil type, detector settings, and GPS. Recordings are stored locally in IndexedDB.
Bundle your dataset as a ZIP (CSV metadata + WAV files) and use the included training script to fine-tune the ONNX model on your own finds.
Native app with NPU acceleration. Download APK
Compose Multiplatform app for macOS, Windows, and Linux. Build from source.
Open /app/ in Safari, then tap Share → Add to Home Screen to install as a standalone app.
The full app runs in any modern browser — no install required. Microphone access needed for live detection.
The easiest way to connect a phone or computer to a detector is from the audio out using a microphone attenuator cable. It looks like a standard audio cord, but has an inline resistor because the headphone out is too strong a signal for most microphone inputs. The “camera” end of this cable is the metal detector.
Most phones also need a 3.5 TRRS to USB-C adapter between the attenuator cable and the device.
Questions? Opening an Issue or Discussion on GitHub is the best place to start. Ask anything.