Avatartarus build a character

TarTalk

Make your avatar talk. The mouth is fifteen vector shapes blended in real time — not a video, not a filter. Nothing you type is uploaded.

Say something

Uses your browser's own speech engine. The text stays on this device — there is no server behind this box.

0/300

Piper voice

Pre-rendered with Piper (en_US-amy-medium) — the same voice engine we use for narration. Fixed phrases, so there's no synthesis endpoint to attack.

Download a video

Records the animation in your browser and saves it straight to your device.

Why the typed-text video is silent: no browser exposes speechSynthesis output as a capturable audio stream, so there is nothing to mux in. The Piper clips record with sound because they're real audio files.