Voice cloning tools can now create a synthetic version of a person’s voice from a recording. Used well, that is helpful: a business can keep a consistent narrator across training videos, a creator can produce audio in less time, and someone who has lost their voice can speak in something close to their own. Used badly, it can mislead listeners, damage reputations or enable fraud.
In this guide you will learn the consent and legal basics of voice cloning in plain language: whose permission you need, what a good consent record looks like, which legal topics tend to come up, how to disclose synthetic audio, and how to set safe internal rules. This is general information, not legal advice. Laws differ by country and change, so use this guide to ask better questions of a qualified adviser, not to replace one.
Key takeaways
- Only clone a voice with the clear, informed, written permission of the person it belongs to, and say exactly what the clone may be used for.
- A voice can be tied up with several legal areas at once, including privacy, personality or publicity rights, contract, copyright and consumer protection; the rules vary by place.
- Consent should be specific, recorded, time-limited where possible and easy to withdraw.
- Be open with listeners when audio is synthetic, especially when it could be mistaken for a real person speaking.
- Never use a cloned voice to impersonate, deceive or pressure anyone, and put security controls around voice samples and models.
What voice cloning is
Voice cloning uses a trained model to produce speech that sounds like a particular person. The tool is given recordings of the voice, learns its characteristics, and can then turn typed text into audio that resembles that speaker. Some tools need a long, clean sample; others work with much less. The quality, and the risk, vary by tool.
Cloning versus ordinary text to speech
Standard text to speech uses a generic synthetic voice that does not belong to a specific real person. Voice cloning is different because the output is meant to sound like an identifiable individual. That identifiability is what raises the consent and legal questions.
Related terms
- Voice sample. The recording used to create the clone.
- Voice model. The trained representation of the voice that is stored and reused.
- Synthetic audio. Any speech produced by a model rather than recorded from a person.
- Voice conversion. Changing one person’s recorded speech so it sounds like another voice.
Why consent matters
A voice is part of a person’s identity. People recognise one another by voice, trust voices on the phone, and attach reputation to how they sound. Using someone’s voice without permission can feel like using their face or signature without asking.
What can go wrong without consent
- Deception. Listeners believe a real person said something they did not.
- Reputational harm. Words are put in someone’s mouth that damage how others see them.
- Fraud and scams. A familiar voice is used to pressure someone into sending money or sharing information.
- Loss of control. A person cannot stop a model of their voice being used once it exists and is shared.
- Disputes. A performer or employee later claims they never agreed to the use.
The business case for doing it properly
Clear consent protects the person whose voice is used and also protects you. A signed, specific agreement is the best evidence that you acted responsibly if anyone later questions the work.
Whose consent you need
Consent comes from the person whose voice is being cloned. That sounds simple, but real situations are not always simple.
| Situation | Whose consent to seek | Notes |
|---|---|---|
| Cloning your own voice | Yours | Still read the tool’s terms about how your sample and model are stored and used. |
| Cloning an employee’s voice for company use | The employee, in writing | Employment alone does not necessarily cover a synthetic copy of their voice. Take advice on contracts. |
| Cloning a freelancer or voice actor | The performer, under a specific agreement | Agreements should address synthetic use separately from a recorded session. |
| Cloning a customer or caller | Avoid unless there is a clear, lawful reason and explicit consent | High privacy risk. |
| Cloning a public figure or celebrity | The individual or their authorised representatives | Being well known does not remove their rights. Avoid. |
| Cloning someone who has died | Seek advice; rights and who holds them vary by place | Sensitive; may involve the estate. |
| Cloning a minor | A parent or guardian, and extra care | Strongly consider not doing it. |
Recording permission is not cloning permission
Someone agreeing to be recorded for a podcast or a video has not necessarily agreed to a model that can say new things in their voice. Treat these as different permissions, and ask for the second one explicitly.
Group and shared recordings
If a recording contains several voices, such as a meeting or interview, each speaker whose voice you want to clone needs to agree. Do not train a model from mixed audio and hope it picks out only the person who consented.
What good consent looks like
Good consent is informed, specific and documented. A vague line in a long contract is a weak foundation.
Qualities of valid consent
- Informed. The person understands what voice cloning is, what the clone can do and what risks exist.
- Specific. It names the purposes, products, languages and channels where the voice may be used.
- Freely given. The person is not pressured, and refusing has no unfair penalty. Be careful with employees, where there is a power imbalance.
- Documented. It is in writing, dated, and stored safely.
- Time-bounded. It has a defined period or a review date where practical.
- Revocable. There is a clear way to withdraw, and a stated plan for what happens to the model when they do.
What to put in a consent record
| Element | What it should say |
|---|---|
| Identity | Who is giving consent and who is receiving it. |
| Purpose | Exactly what the cloned voice will be used for. |
| Exclusions | What it will not be used for, such as political, adult, medical or endorsement content. |
| Scope | Languages, channels, regions and projects covered. |
| Duration | How long the permission lasts and what happens at the end. |
| Compensation | Any payment or other agreed arrangement, if relevant. |
| Storage and security | Where samples and models are kept, and who can access them. |
| Review and approval | Whether the person approves scripts or outputs. |
| Withdrawal | How to withdraw consent and how deletion will be handled. |
| Disclosure | How listeners will be told the audio is synthetic. |
Consent is not forever by default
Do not assume consent given once covers every future use. If you want to use the voice for a new purpose, ask again. Revisit the arrangement when the project ends, when the person changes role or leaves, or when the tool or provider changes.
Legal areas that may apply
There is no single law called “the voice cloning law” that applies everywhere. Instead, several established areas of law may be relevant, and their details differ by country and region. The list below is a map of topics to raise with a qualified adviser, not a statement of what the law is in your location.
| Area | Why it may matter | Question to ask |
|---|---|---|
| Privacy and data protection | A voice recording can be personal data, and in some places a voice model used to identify someone may attract extra protection. | What is our lawful basis, and how do we handle storage, access and deletion? |
| Personality, publicity or image rights | Some places give people rights over commercial use of their identity, which may include voice. | Does this use trade on a recognisable person’s identity? |
| Contract and employment | Existing agreements may not cover synthetic voice use. | Does the contract expressly permit this, or do we need a new one? |
| Copyright and performers’ rights | Recordings and performances can carry rights separate from the voice itself. | Do we have the right to use the source recordings for training? |
| Consumer protection and advertising | Misleading audio in marketing can breach rules on fair advertising. | Could listeners be misled about who is speaking or endorsing? |
| Fraud, impersonation and defamation | Using a cloned voice to deceive or damage someone may be unlawful. | Is there any risk this could be taken as the real person? |
| Sector rules | Regulated fields such as finance, healthcare or broadcasting may add requirements. | Do industry rules apply to our use? |
| Emerging AI rules | Some jurisdictions are introducing rules on labelling synthetic content. | What disclosure duties apply where our audience is? |
Why location matters
A voice used in one country may be heard in many. Rules can depend on where the speaker lives, where your business operates and where the audience is. When your content is international, take advice that covers the places that matter.
Keep up to date
This area is changing. Review your approach periodically and when you enter a new market, rather than relying on a single check made years ago.
Use cases by risk level
Not every use of a cloned voice carries the same risk. Sorting them helps you decide where to apply the strictest controls.
| Risk level | Examples | Approach |
|---|---|---|
| Lower | Your own voice for your own content; an employee narrating internal training with written consent and clear limits. | Document consent, label the audio and secure the model. |
| Medium | A contracted performer’s voice for customer-facing videos; localisation of a speaker’s voice into other languages. | Specific contract, approval rights, disclosure and legal review. |
| Higher | Customer service calls that sound like a named person; endorsements or testimonials; voices of the deceased. | Take specialist advice. Consider whether a generic synthetic voice would do. |
| Not acceptable | Impersonating someone without permission; fake statements from real people; scams; political deception. | Do not do it. |
Ask whether you need a clone at all
Often a generic synthetic voice or a human recording will do the job with far less risk. A clone is worth the extra care only when the identity of the specific voice truly matters to the project.
Endorsements need extra caution
A cloned voice saying that a real person recommends a product is a particular concern. Do not generate endorsements, quotes or testimonials in someone’s voice unless they have approved the exact wording and the use.
Disclosing synthetic audio
Telling listeners that audio is synthetic builds trust and reduces the risk of misleading anyone. In some places there may also be a legal duty to do so, and platforms often have their own rules.
When to disclose
- When a listener could reasonably believe a real person is speaking live or recorded.
- When the voice belongs to an identifiable person.
- In advertising, news-like content or anything that could influence a decision.
- On phone or voice assistants that interact directly with the public.
How to disclose well
- Say it plainly. A short statement such as “This audio was generated with a synthetic voice” works better than buried fine print.
- Put it where people will notice. At the start of the audio, in the description, and on the page where it is hosted.
- Be consistent. Use the same wording across your channels.
- Respect the speaker. Check that the speaker is comfortable with how it is described.
- Check platform rules. Hosting platforms may require labels for synthetic or altered media.
Interactive voices
If a synthetic voice talks with customers, tell them at the start that they are speaking to an automated system and offer a route to a person. The planning points in how to plan an AI customer support chatbot apply equally to voice.
Security and misuse
A voice model is a sensitive asset. If it is stolen or misused, the person whose voice it is can be harmed, and you may be responsible.
Protect the samples and models
- Limit who can access samples and models, and keep a log of use.
- Store them in an approved, secure location, not on personal devices or shared drives open to everyone.
- Do not upload voice samples to a service until you understand how it stores and uses them.
- Delete samples and models when consent ends, and record that you did so.
The broader principles are in how to protect company data when using AI tools.
Reduce the risk of impersonation scams
Because voices can be imitated, do not treat a familiar voice alone as proof of identity, particularly for payments or access requests. Use a second check, such as calling back on a known number or using an agreed code word.
Watch for misuse inside your team
Set a rule that cloned voices are used only for the approved purposes, require approval for new uses, and make it easy to report concerns. Keep scripts that use a cloned voice under review.
Questions for tool providers
Before choosing a voice cloning product, ask the provider clear questions and read the terms. The approach in how to choose an AI tool applies here, with a few voice-specific additions.
| Topic | Question |
|---|---|
| Consent checks | Does the tool require proof that the speaker has agreed to the cloning? |
| Data use | Are uploaded samples used to train the provider’s own models? |
| Storage and deletion | Where are samples and models stored, and can they be deleted on request? |
| Ownership and licence | Who holds rights in the voice model and the generated audio? |
| Misuse controls | What safeguards exist against impersonation and abuse? |
| Labelling | Does the tool support marking audio as synthetic? |
| Access control | Can we restrict who on our team can use a given voice? |
| Exit | What happens to our voices and data if we stop using the service? |
Read the terms yourself
Terms of service can include licences over content you upload. Read them before uploading anyone’s voice, and check them against what you promised the speaker in the consent agreement.
Setting internal rules
If more than one person in your organisation may use voice cloning, write down simple rules. A short, clear policy is better than assumptions.
What the rules should cover
- Who may approve a new voice clone and the purposes allowed.
- The consent process and where records are kept.
- Banned uses, such as impersonation, endorsements without approval and anything deceptive.
- Disclosure requirements for audio shared externally.
- Security rules for samples and models.
- How a person can withdraw consent and what happens next.
- Who to contact with concerns or incidents.
You can fold these into a wider document; see how to write an AI use policy for your team.
Train the people involved
Make sure staff who handle audio understand why consent matters and how to spot misuse. A short briefing at the start of a project is cheap compared with fixing a dispute later.
Templates and checklist
Plain-language consent statement (starting point)
“I, [name], agree that [organisation] may create a synthetic version of my voice using recordings I provide. It may be used only for [purpose, products, languages, channels] until [date or event]. It will not be used for [exclusions], and nothing will be generated in my voice that implies I endorse a product or hold a view without my approval. Audio will be labelled as synthetic. My recordings and the model will be stored securely, accessible only to [roles]. I can withdraw this consent at any time by contacting [contact], and [organisation] will stop using and delete the model within [period] and confirm in writing. Signed and dated.”
Have a qualified adviser review any consent wording before you use it. This template is a starting point only.
Pre-project checklist
- Is a cloned voice really necessary, or would a generic voice do?
- Do I have specific, written, informed consent from the voice owner?
- Have the purposes, exclusions, duration and withdrawal route been recorded?
- Have I taken advice on the legal topics that apply where I operate and where my audience is?
- Do I have the rights to the source recordings?
- Have I read the provider’s terms on data use, storage and deletion?
- Is there a plan to label audio as synthetic?
- Are samples and models stored securely with restricted access?
- Is there a named person who approves each new use?
Frequently asked questions
Can I clone my own voice without any worries?
It is much lower risk because you are the person consenting, but still read the tool’s terms on how your sample and model are stored and used, and keep the model secure so others cannot misuse it.
Is it legal to clone someone’s voice?
It depends on who the person is, how you use the clone and where you are. Several areas of law can apply. This is general information, not legal advice, so speak to a qualified adviser about your situation.
Do I need consent if I only use the voice for a small project?
Yes, as a matter of good practice. The size of the project does not change the fact that the voice belongs to a person who should decide how it is used.
Can I use a voice from a video I found online?
Being publicly available does not mean you have permission to clone it. The speaker’s consent and any rights in the recording still matter. Avoid it.
What if an employee leaves?
Your consent agreement should say what happens. Plan for it in advance: usually that means ending use and deleting the model, unless the person has agreed otherwise in writing.
Do I have to label synthetic audio?
Requirements differ by place and platform, and they are evolving. Even where it is not required, labelling is good practice and protects trust. Check the rules that apply to you.
Can a cloned voice be used for customer calls?
It can, but tell callers they are speaking with an automated system and provide a route to a person. Avoid making the voice sound like a real, named individual unless that person has agreed and the use has been reviewed.
Conclusion
Voice cloning is a powerful tool that comes with a simple rule: the voice belongs to the person, so they decide. Get specific, informed, written consent, spell out the purposes and exclusions, store models securely, and tell listeners when audio is synthetic. Think about the several areas of law that can apply, and take advice for anything customer-facing, commercial or international, remembering that this guide is general information, not legal advice.
Build the habit early with a short internal policy and a consent template reviewed by a professional. When you are weighing products, use the questions above together with how to choose an AI tool, and protect everyone involved with the practices in how to protect company data when using AI tools.
