Voice cloning and consent

The question is about the voice and the use

Voice cloning makes generated speech resemble a particular person's voice. The question is not only whether a tool can make a convincing sound: it is whether the voice is identifiable, who authorized its creation, and what the resulting audio will communicate. A person's agreement to provide a recording may not answer every question about later uses, audiences, or distribution.

Rules vary by place and use. For example, Tennessee's ELVIS Act has been in force since 1 July 2024 and adds voice as a protected personal right. That is one jurisdiction's example, not a rule for every place. Other laws may apply to a particular use, and a service's terms can set additional conditions.

Check permission, purpose and reach

Start with the speaker: is it your own voice, or can you show that the person whose voice is identifiable agreed to the specific plan? A useful permission record describes making the voice model and its intended uses. Microsoft's guidance for its custom voice service calls for explicit written permission and recommends explaining where and how people may hear the voice, and what it may be made to say.

Next, consider the use and its audience. A private accessibility narration differs from a public advertisement, an endorsement, or audio that could make listeners believe the person said something they did not. Check whether the permission covers the audience, channels, duration, edits, and sharing you have in mind. Review the provider's terms too: a platform may require evidence or set limits regardless of what local law requires.

A simple rule of thumb

Proceed only when you can identify whose voice is being modeled, have permission that clearly covers both creating the clone and the planned use, and have checked the rules and provider terms relevant to that use. If the permission is vague, the recording's source is unclear, or the intended use changes, pause and resolve that gap before generating or sharing the audio.

Even with permission, think about how a listener will understand the result. If the audio could be mistaken for a real recording, make its synthetic nature clear in the context where it is heard. For a request that arrives as a familiar voice asking for money or sensitive information, verify it through a separate, known contact route; a voice alone is not confirmation.

Three examples

You want to make reminders in your own voice. You can identify the speaker and the purpose is narrow. Check that the provider permits this use, then decide whether anyone else will hear the recordings and whether that changes what you want to share.

A voice actor records samples for a project. Permission for recording alone may not describe every later use. Before making a clone, set out the intended contexts and what the model may say, then check that the actor's permission covers those uses and the provider's terms.

You find a relative's voice in an old family recording and want a clone for a public message. The recording does not by itself establish permission for this new use. If you cannot establish the relevant permission or check the rules where the audio will be used, do not proceed on the assumption that access to the sample settles the question.