Voice cloning for brands: consent and best practices

A cloned voice can give a brand something that is hard to get any other way: the same recognisable voice in every ad, tutorial, phone prompt and product video, available whenever the script changes. It also involves a real person's voice, which means the work starts with permission, not with a recording. This guide covers how brands can approach voice cloning responsibly: getting consent, putting it in writing, defining scope, planning for the day it ends, and keeping the result consistent.

This article is practical guidance, not legal advice. Rules on voice, likeness and personal data differ between countries. Before you clone anyone's voice for commercial use, talk to your own counsel.

Whose voice is it?

Brands typically clone one of three kinds of voices:

• A founder or spokesperson who already represents the brand.
• An employee, such as a trainer or a support lead, whose voice customers know.
• A hired voice talent engaged specifically to become the brand voice.

In every case, the voice belongs to the person, not to the company. Even if the voice talent was paid, and even if the employee recorded the samples during work hours, you need their explicit agreement to clone their voice and to use the clone in the ways you plan.

There is also a fourth option that needs no one's consent: a voice that belongs to no one. With voice design you can create an original voice from a description, or use a voice from the public library. If what you need is consistency rather than a specific person, this is often the simpler path.

Get consent, and get it in writing

Vocla's rules are simple: every cloned voice must be your own voice, or one whose owner has given explicit, written permission for how you will use it. Cloning public figures, or anyone without consent, is prohibited. Our voice consent policy sets out the details.

Good consent is informed. Before the person agrees, explain in plain language:

• that an AI model will be built from their recordings,
• that it will be able to say things they never recorded,
• where the audio will be used, and for how long,
• who in the company can generate with it,
• how they can withdraw.

Then write it down. A short written agreement, reviewed by your counsel, is far better than a verbal yes in a meeting.

What the agreement should cover

Your lawyer will shape the final document, but brands usually want to settle these points:

1. Scope of use: which channels (ads, social, IVR, e-learning, internal training), which languages and dialects, and which markets.
2. Content limits: topics the voice will never be used for, such as political messages, competitor comparisons or anything the person would find objectionable. Let the person add their own.
3. Approval: whether the person reviews scripts, a category of scripts, or nothing after the initial agreement.
4. Duration: a fixed term, renewable, rather than "forever".
5. Compensation: how the person is paid, if at all, and whether usage affects it.
6. Ending the arrangement: what happens if the person leaves the company or withdraws consent, and what happens to content already published.
7. Deletion: a commitment to delete the cloned voice when the agreement ends.

Plan for the end from the start

Employees leave, contracts end and people change their minds. Decide now what will happen:

• New content stops on the agreed date. On Vocla you can delete a cloned voice at any time, and the voice model built from the recordings is deleted with it.
• Existing content: agree whether published ads and videos stay up, are phased out, or are replaced.
• A successor voice: if the brand voice is central to your identity, think about how you will transition, for example by introducing a designed voice gradually.

Be open with your audience

Disclosure is a matter of trust, and sometimes of platform rules. Where listeners would naturally assume a human is speaking live, such as in a phone call or a video where the person appears to address viewers directly, consider telling them that the voice is AI-generated. Several platforms have their own disclosure requirements for realistic synthetic content; read them for each channel you use.

Never use a cloned voice to make a person appear to say something they did not agree to say, and never use it to impersonate anyone else.

Recording samples that produce a good clone

A clone is only as good as its recordings:

• Record in a quiet room with soft furnishings. Echo and background noise carry into the model. If you only have imperfect recordings, the voice cleaner can remove background noise first.
• Record in the dialect you want the brand voice to use. A clone learns how the person speaks, so a Saudi speaker recorded in Saudi Arabic stays Saudi.
• Vary the material: some calm sentences, some energetic ones, questions, numbers. Vocla accepts up to 20 clips; ten seconds of clean speech is enough to start, and a minute or two gives the best likeness.
• Keep the same microphone and distance across clips.

Keeping the brand voice consistent

Once the clone exists, consistency comes from process:

• One owner: name a person responsible for the voice, its settings and the scripts that use it.
• A pronunciation list: add your brand name, product names and recurring terms to the pronunciation dictionary, so every generation says them the same way.
• Script guidelines: a one-page style guide with tone, typical phrases and words to avoid.
• Seats and access: a cloned voice on Vocla is private to the workspace that made it and never appears in the public library. Only give workspace access to people who are allowed to generate with the voice.

Which plans include cloning

Voice cloning is included from the Starter plan, at $4.99 a month with one cloned voice. Professional, at $29 a month, includes five cloned voices and three seats, which suits a brand with a main voice and a few regional or campaign variants. Larger teams can talk to us about Enterprise. All paid plans include commercial rights for generated audio, as long as you have the right to use the voice itself. See pricing for current details.

A short checklist

1. Decide whether you need a specific person's voice, or whether a designed or library voice would do.
2. Explain the process to the person and get informed consent.
3. Put scope, limits, duration, compensation and deletion in a written agreement reviewed by counsel.
4. Record clean, varied samples in the right dialect.
5. Set an owner, a pronunciation list and access rules.
6. Disclose AI narration where your audience would assume a human.
7. Delete the voice when the agreement ends.

For the wider picture on consent and AI voices, read Consent first: the ethics of using AI voices.