Wake Word Benchmarks

Score a wake word model against a suite of augmented test clips - clean, noisy, far-field, and reverberant - and get recall plus false-accepts-per-hour metrics that correlate with real-room performance. Runs are shared with your OpenWakeWord account.

Realistic test conditions

Positive clips are synthesized across many voices and augmented with real background noise and room impulse responses, then scored alongside long stretches of real conversation to measure false activations.

Compare versions

Keep a history of benchmark runs so you can compare model versions and pick the build with the best recall at an acceptable false-activation rate before you ship it.

Browse benchmarked models
Create a voice for testing

This page is an app that requires JavaScript. Enable JavaScript to use the training center and interactive library.