A voice clone starts with a recording of a real person, so the first thing a provider asks a business is whose voice it is. What a provider requires can be read off its own published rules, while what a law allows is a question for a lawyer, and the two are not the same thing. Nothing here is legal advice; it reports what each provider publishes.
Providers ask for four kinds of proof, as of November 2026: a spoken challenge phrase, a signed agreement, a recorded statement compared against the voice, or a confirmation the customer gives. Bland Speech uses the last for every clone and adds a consent and rights review for professional ones. A rule that names a place is the provider's own restriction, not a statement of law.
Is voice cloning legal for a business?
Whether a business may clone a particular voice is a legal question that no text to speech provider answers for it, because a provider's rules say what the provider requires and a law says what is allowed.
Several providers start in the same place. Bland says a customer confirms every voice is theirs to use. ElevenLabs bars a customer from providing input or creating output without all the rights necessary, and Inworld's clone form asks a customer to confirm rights to the voice before it continues. Sarvam says plainly that voice cloning requires consent, and Typecast bans cloning another person without prior, explicit, and verifiable consent.
A provider's check is also its own rule and not a ruling. ElevenLabs says its verification cannot guarantee that a recording truly belongs to the requester. A business that clones an employee, a spokesperson, or an actor can treat the provider's steps as a starting point, but what a written agreement with the speaker should say, and whether it is sufficient, are questions for its counsel.
What does each provider require before a business can clone a voice?
Bland Speech has a customer confirm that every voice is theirs to use and sends professional clones through a consent and rights review, while Speechify makes the speaker read a spoken challenge phrase, ElevenLabs runs a voice captcha, and Azure compares a recorded statement against the training audio.
What these rows mean
- What the customer provides: the consent step each company publishes. A provider is in the table when it publishes one.
- Who checks it: the party that confirms the step. Speechify, ElevenLabs, OpenAI, Azure, and Google require a recorded step, and Speechify, ElevenLabs, and Microsoft say how it is checked. Typecast, Inworld, Sarvam, and Murf rely on what the customer states, and Typecast may ask for evidence later.
- Bland Speech: Bland says a customer confirms every voice is theirs to use and that professional clones require passing Bland's consent and rights review for the voice being cloned.
A check the provider runs
Speechify's consent is a spoken challenge phrase the speaker reads, checked and kept as the consent record, with no path that skips it. ElevenLabs' professional clone verification has the voice owner read captcha lines aloud, and manual verification exists only as a fallback. Microsoft says it reserves the right to use speaker recognition on a recorded acknowledgement statement for Azure voices, and says the voice signature in that statement is compared with the training recording. Google's Instant Custom Voice required a recorded consent statement before it closed to new requests.
A promise the customer makes
Typecast, Inworld, Sarvam, Murf, and Bland Speech rely on a confirmation or warranty from the customer. Bland's confirmation on an instant clone is a statement from the customer, as Inworld's and Sarvam's are, and Bland adds its consent and rights review to professional clones.
A gate on who may clone at all
Some providers limit cloning before consent comes up. Azure's custom voice is available by registration only, for approved use cases, and OpenAI limits custom voices made from recordings to eligible customers. Respeecher builds custom voices as projects with its own team.
What changes when the voice belongs to someone else?
The rules get stricter when the voice is a person's and not the customer's own: Azure asks for written permission from voice talent, OpenAI asks for the voice actor's consent recording, Descript requires recorded authorization from the person, and Respeecher requires a signed agreement.
Azure's rule is a warranty, which means the customer states it has explicit written permission from voice talent before creating a voice model. OpenAI's custom voices need a recording in which the voice actor gives consent to create a likeness of their voice. Descript says it requires explicit recorded authorization from the person whose voice will be used, and says it cannot create voice clones for a deceased individual.
Bland's professional review covers this case: Bland says training a professional voice requires passing its consent and rights review for the voice being cloned. Clones made on Bland Speech are private to the customer's organization.
What happens to a voice sample after it is uploaded?
A sample and the clone made from it follow each provider's data rules, and they differ: Typecast may train on submitted voice data, Descript uses training audio to improve its service, and Inworld's zero data retention does not cover clone samples.
Others run the other way. MiniMax deletes a cloned voice that goes unused for 7 days, Azure deletes a lite voice and its recordings within 90 days unless full access is granted, and Mistral attaches a retention notice to saved voices, with 30 days as the default and 365 as the maximum. Grok says custom voices are scoped to a team and never available to other users, and Bland says clones are private to the customer's organization. Fish Audio is the one to read closely: a professional voice a customer releases can earn a share when other creators pay to use it, which matters to the speaker whose voice it is.
What should a business settle with a lawyer before cloning a voice?
A lawyer will work through at least five topics before the first sample is recorded: consent in writing, the uses allowed, withdrawal, how callers are told, and where the speaker and callers are.
- Who signs. What a written agreement with the speaker names, and what it covers that a provider's consent check does not.
- What it is for. Which calls, channels, and duration the agreement covers, set beside Azure's approval, which is limited to the use cases registered, and OpenAI's custom voices, which are limited to eligible customers.
- How it ends. What happens to a sample and a clone on withdrawal, set beside each provider's deletion and retention rules.
- How callers are told. Whether and how callers learn the voice is synthetic, given that OpenAI requires a clear disclosure to end users that a TTS voice is AI-generated.
- Where people are. Which places the speaker and callers are in, given that Grok's custom voices are available only in the United States and not in Illinois, and that Speechify refuses cloning requests by IP address from jurisdictions where it is unavailable.
Bland Speech fits a business that wants clones kept private to its own organization, with a customer's confirmation on every voice and Bland's review on professional ones. The guide to protecting a cloned voice covers access, and the best voice cloning with consent verification ranks providers on the check alone.
Bland Speech is also covered in instant vs professional voice cloning, in the voice cloning comparison hub, and in the wider text to speech comparison. Bland Speech is where a clone starts.
Frequently asked questions
Is voice cloning legal?
A provider cannot say, because its published rules cover what it requires and a law covers what is allowed. Every provider in the table asks for consent or a confirmation of rights, and what a business owes beyond that is a question for its lawyer.
Can a business clone an employee's voice?
A provider will clone a voice when the speaker has consented in the way it requires, such as Speechify's spoken challenge phrase, Azure's written permission from voice talent, or Respeecher's signed agreement. Whether a written agreement with the employee is also needed, and what it says, is a question for counsel.
Can a business clone the voice of someone who has died?
Descript says it cannot create voice clones for a deceased individual. For any other provider, a business asks before assuming either way.
Does a provider's consent check make a clone compliant?
No. A provider's check is its own requirement, and ElevenLabs itself says its verification cannot guarantee that a recording truly belongs to the requester.
Do callers have to be told a cloned voice is AI?
OpenAI requires a clear disclosure to end users that a TTS voice is AI-generated. Whether a law adds more is for a lawyer to say.
Does any provider restrict cloning by location?
Grok's custom voices are available only in the United States, with the exception of Illinois, and Speechify says voice cloning is not available in some jurisdictions.
Who can use a clone made with Bland Speech?
A clone made with Bland Speech is private to the customer's organization, and the customer confirms every voice is theirs to use. Professional clones also pass Bland's consent and rights review.
