Audio to Talking Avatar

Experimental

This is an experiment. The results are still rough, and it may change or be removed without notice.

Makes a talking avatar video from an audio file or a recording, entirely in this browser: your audio, the 3D model you upload and the finished video never leave your device, and the only downloads are the tool’s own files (the 3D engine and two preset models). The mouth follows the sound rather than the words, so it reads as speech without matching every syllable. Pick a 3D model — one of two CC0 VRoid samples or your own VRM file — or the light 2D avatar, then save an MP4 or WebM video with the audio.

How to use it

Choose a voice file or record one. A file is MP3 or WAV, up to 20MB and 60 seconds, and a recording ends by itself at 60 seconds.
Pick an avatar and a background.
Play the preview and watch the mouth movement.
Pick the format to save in, save the video, and download it.

Good to know

The mouth follows the sound, not the words. It does not match every syllable.
Saving runs in real time: it takes as long as the video is.
A reload discards the current work. Download a saved video straight away.
In a browser that cannot save video, the preview still works.

Common questions

Is my voice sent anywhere?

No. The voice, the model, and the finished video are handled on this device only, and the only things downloaded are the files of the tool itself. Why the tools on this site are built this way is in What processing in the browser means.

What if the sound is quiet?

When almost nothing can be heard the mouth does not move, and the page says so. You can still save, but trying again with a louder file works better.