Check an AI voice

The host’s voice is made by AI in advance, and our server sends it as audio files. Each file carries a mark that says it is AI-generated. When a signing key is configured, the OGG files the website plays are signed by us. The bots’ voices are made on your own device, not on our server — see the end of this page. Upload a clip, or give the address of a clip on this site, and see what it says.

Check a clip

What the mark says

Every OGG or WAV file our server sends carries SYNTHETIC=1, the generator’s name and the IPTC term trainedAlgorithmicMedia (the standard word for AI-made media). The response also carries the header x-content-synthetic: 1. When a signing key is configured, the file also carries the signing time, an audio fingerprint, a key ID and an Ed25519 signature. That signature covers the fingerprint, time, key ID and source-type value. The m4a copies of the host’s lines that the iPhone app plays carry the same text mark, without a signature.

Check it yourself

Read the mark with any audio tool, for example ffprobe -show_entries format_tags:stream_tags clip.ogg. The public key the signature is checked against is at /api/ai-mark/key.

Nothing is kept

The file is checked in memory and forgotten at once: no copy, no record of what was in it. Checks are limited to a few a minute from one address.

Voices made on your own device

The bots’ voices are made by your browser or the app on your device, from voice models it downloads from our server; our server sends the bots’ lines as text and makes no bot audio. A voice made on your device never becomes a file we could mark, so for it the table always carries one line: “Bots and the host are AI; their voices are synthetic.”