Privacy: everything runs in this tab. No audio leaves the page, nothing is uploaded, no analytics.
The only network requests are loading this page's own static files (scripts, models, runtime).
Results go out only when you press Copy / Download.
Not started
Results also show what 1, 2 and 3 frames would each have done. 1.5 s refractory after a detection. A beep plays on detection.
score 0.000last detection: -frames: 0backlog: 0
Mode A - 30 tries
Distance:
Hits 0Missed 0Tries 0/30
Say "hey jarvis" at normal volume, wait for the beep (or ~2 s), repeat. Press Missed when it did not fire.
Switch distance between sets. Each hit/miss records the distance selected at that moment.
Mode B - false-wake soak
00:00False wakes 0Near-misses (peak 0.3-θ) 0
Run 30 min of normal room activity (talking, typing, TV) without saying the wake phrase. Every detection counts as a false wake.
Hidden-tab probe
5-min probe: Hide this tab (switch to another tab or minimise) for 5 min, say "hey jarvis" near the end, come back. 30-min probe: Hide this tab (switch to another tab or minimise) for 30 min, say "hey jarvis" near the end, come back.
Visible fps -Hidden fps -Max gap while hidden - ms
#
hidden at
dur s
frames
fps
max gap ms
detections while hidden
Perf
Inference ms median - / p95 -Audio→score lag ms median - / p95 -Max backlog 0End-of-speech→detection ms median - / p95 -(approx.)