Hey Jarvis test

Jarvis stage S3. Runs entirely on this phone; no audio leaves it (tap-to-talk below uses Chrome's own speech service).

Not listening

score 0.00 · threshold 0.50 · test time 0:00

level – · speed – · backlog – · audio steps 0

Threshold:

Your counts

0heard you
0missed you
0false alarms

Each time it says "heard", tap "Not me" if you didn't say it. Tap "I said it, it missed" when it didn't react.

    The test (about 10 minutes)

    1. Tap Start listening and allow the microphone.
    2. Screen on, 5 minutes: say "Hey Jarvis" about 10 times at normal volume from different distances, and talk normally in between.
    3. Screen off, 5 minutes: lock the phone with this page open, say "Hey Jarvis" about 5 times, then unlock and check the log.
    4. Tap Copy results and paste them into the Jarvis chat.

    Teach it your voice (about 2 minutes)

    Your recordings stay on this phone. It learns from them and keeps a small voice model here.

    1. 20 × "Hey Jarvis": after each beep, say "Hey Jarvis" the way you'd normally say it (about 60 s).
    2. 1 minute of normal talk: talk, read something aloud, or let the TV play. Don't say "Jarvis".
    3. Train: takes a second. Then run the test above again with "My voice" selected.

    Nothing recorded yet.

    Listen with:

    Tap-to-talk

    The fallback when the wake word misses. Tap, then say a command such as "what's waiting" or "option two".

    Nothing heard yet.

    Wake-word models: openWakeWord pre-trained "hey jarvis", melspectrogram and embedding models by David Scripka, CC BY-NC-SA 4.0, used privately. Inference: ONNX Runtime Web.