The voice you cannot borrow

There is a rule in ElevenLabs’ voice-cloning documentation worth pausing on because it draws a narrow, practical line. For its Professional Voice Cloning product, a user can create a Professional Voice Clone only of their own voice. Verification is required. Another person’s consent does not let you create that person’s Professional Voice Clone in your own account. The voice owner creates and verifies it, then can share it privately.

I want to keep the boundary of that claim clear. This is a vendor rule for Professional Voice Cloning. It is not a statement of universal law, and it does not describe every voice-cloning method or every text-to-speech product. The verification process establishes that the account holder has met the product’s own qualification requirement. It does not establish legal ownership of a voice, guarantee perfect resemblance, or settle every question about identity.

What the rule leaves open is more interesting. Permission to use a particular voice is a different decision from permission for that voice to speak particular words in particular settings. The first concerns access to a voice model. The second concerns representation, context, and responsibility. I think the second permission deserves its own question every time.

Consider a clearly fictional case. A narrator creates and verifies their own Professional Voice Clone, then shares it with a publisher to read one audiobook. Later, the publisher wants that same model to read a political advertisement or a different book the narrator has not reviewed. The model may resemble the narrator’s voice, and the product verification may have been completed correctly. Neither fact answers whether the narrator agreed to that new use.

The documented check answers a product-access question. It does not answer what the voice may be used to say. That is my ethical distinction, not a claim made by the vendor. Consent should name the context, the material, the audience, and whether later uses require another agreement. A person can authorize one reading without authorizing every future sentence.

The practical lesson is simple. Ask two questions instead of one: has the product’s voice qualification been completed, and has the person agreed to this particular use? The first may be documented by a setting. The second remains a human permission, and the documentation cannot answer it for you.