1. The rule
Before a voice model is trained, you must record your agreement to this statement, word for word. The studio shows it to you and will not train — or accept a dataset — without it:
“I confirm this recording is my own voice, or that I have the written permission of the person whose voice it is, and that I will not use the trained model to impersonate anyone without their consent. I understand I can delete this voice data at any time.”
Only one voice is trained at a time in your account, and each voice profile carries its own consent record: agreeing for one voice does not cover another.
2. Whose voice may I use?
- Your own. No further permission is needed.
- Somebody else's, with their written permission. A message granting it, an agreement, a signed release — keep it; you may be asked for it.
- Not a public figure's voice to make it appear they said or sang something, and not a voice taken from a recording without permission.
3. What we record
When you attest, we store the version of this statement, the exact wording you agreed to, and the time you agreed. We do not store audio of the attestation.
4. Deleting voice data
Deleting a voice profile removes the dataset recordings, every trained model for that profile, and the profile itself — files and records together. You cannot delete a profile while a training job is running; wait for it to finish. Deletion cannot be undone.
5. If your voice was used without permission
Use the takedown process. We take unauthorised voice use seriously and act on it quickly.