AI voice licensing is permission from a real person to use their voice or an authorized voice model for a defined AI use. It should say who can use it, for what, where, for how long and under which approval rules.
Recording a voice and licensing it are different acts. A recording creates source material. A licence defines what another party may do with it. Possession of an audio file or access to a voice model is not, by itself, permission to use the person in a campaign.
A defensible record connects the identified person, the signed licence, the authorized voice asset and each generated output. Without that connection, a buyer may have a contract but still be unable to show that the specific file in front of them was made under it.
Before recording, use the voice cloning checklist. Before publishing, use the licensing checklist.
No. A recording is source material. A licence is permission for a specified party to use the voice within agreed limits.
No. A licence grants limited use rights. It does not sell the person or make their identity a reusable stock asset.
Future generation should stop, and the agreement should state what happens to outputs already published.