Heed Wake Word

Real-time wake-word detection in your browser. All audio stays on this device. Nothing is sent anywhere. Train your own wake word in Colab →

Loading model…

0.0 threshold: - 1.0
raw probability- smoothed (EMA)- energy gate- inference latency (median, ms)-
Triggers:
(none yet)

Note: the energy gate shown here is a lightweight time-domain approximation of the on-device Python gate, so under steady background noise it can stay "passed". It only decides whether to skip computation and does not change what triggers.

Test your own model

Drop your wake.onnx and wake.json here (the files from heed export), or . It runs entirely in your browser, nothing is uploaded. ONNX here (.tflite is for the mobile delegates).