Voiceprint

Speaker recognition that never leaves your device. Runs AntResearch/AntSpeaker (MECT-B2) in your browser.

Starting
Audio is analysed inside this tab and is never uploaded. A voice is reduced to a 192-number voiceprint, saved only in this browser if you choose to.
1

Enroll a voice

Record three short clips of the same person, about four seconds each. Any words work.

2

Try to unlock

Enroll a voice first
Then speak, upload a clip or try a sample.
0.45

    Finds where people speak, fingerprints every 1.5 s of speech and groups the fingerprints by voice. Known voices from Voice ID get named automatically.

    People you know

      Each person needs one clip of a few seconds. Adding more clips to the same name sharpens their voiceprint.

      Who is this?

      Add people, then play someone a clip.
        0.45