Best AI Voice Cloning Software in 2026

✓ Docs-verified · Sep 1, 2026◌ Not audio-tested△ What we could not verify

Published September 2, 2026 · Every price, access gate and consent requirement below was read from the vendor’s own documentation on September 1, 2026 and carries that check date. This page ranks documented capability and controls, not how any cloned voice sounds — no listening test has been run.

Status correction, September 2, 2026. (correction record) Resemble AI appears in the ranking below and no longer sells voice cloning. The vendor names it directly: it does not sell “text-to-speech, voice cloning, speech-to-speech conversion, or voice design as commercial products”. Its open-source models remain published, and the vendor now calls them “research artifacts rather than supported products” carrying “no SLA, paid tier, or enterprise support”. Resemble AI’s own machine-readable description, dated August 13, 2026 and read on September 2, 2026, states verbatim: “Resemble AI does not sell voice generation products and is not accepting new voice customers. Voice continues as open research only.” The full review carries the correction notice and the surrounding quotations.

Quick verdict

Cloning access in this market runs from $5 a month to a $1,000-a-month plan to “contact sales” — and price tracks almost nothing else. Cartesia at $5/month and ElevenLabs at $6/month are the cheapest self-serve routes. Resemble AI documents its cloning API as requiring a $1,000/month plan. Amazon Polly, Azure and OpenAI publish no self-serve route at all.

The axis that should decide this, and usually does not, is consent. Some providers verify it by recording before a voice can exist; others take a warranty from you and check nothing. Those are very different products wearing the same label, and the difference is not reflected in the price.

The finding a buyer most needs: the cheapest documented cloning route in this comparison — MiniMax’s pay-as-you-go rapid cloning, with no subscription required — is also the one where we found no voice-owner consent requirement published anywhere in its developer documentation. That is a searched-and-not-found result across both verification passes, not an oversight.

Ranking basis: documented capability, access and consent controls only, checked September 1, 2026 — no accounts created, no voices cloned, no listening test, no numeric scores (how we verify).

The ranking

The order weighs three documented properties: whether cloning is reachable without a sales conversation, what it costs, and how seriously consent is enforced. Enforcement is weighted heavily — a cheap cloning tool with no consent gate is not a better product, it is a riskier one.

Ten providers ranked on documented cloning capability — verified September 1, 2026
#ProviderCheapest documented routeConsent enforcement
1ElevenLabs$6/month Starter (instant); $22/month Creator (professional)Strong. Professional cloning is your own voice only, with identity verification
2Speechify$10/month API StarterStrongest technical enforcement documented. Machine-verified by recording; no unattended path exists
3Cartesia$5/month Pro (instant); $49/month Startup (professional)Contractual only. A warranty you give; the cloning API has no consent field
4MiniMaxPay-as-you-go, no subscription — the cheapest documented route hereNone found. Identity-gated on the developer; no voice-owner requirement published
5Resemble AICannot be reconciled: a changelog says “first clone free, then $2 each”; the docs require the $1,000/month Business planDiscretionary. Resemble “may require consent”, and consent “needs to be verbal”
6Google Cloud Text-to-SpeechNo plan; $60 per 1M characters, but access is allow-listed via salesMandatory and scripted. A recorded consent statement submitted with the audio
7Azure Speech in Foundry ToolsCannot be priced. Limited Access, Microsoft-managed customers onlyMost heavily enforced here. Mandatory recorded statement, backed by biometric matching
8OpenAINo price published. Custom voices limited to eligible customers via salesStructurally enforced. Two separate recordings required
9Murf AINot self-serve at any tier — an enterprise engagementStrong on paper. Explicit written consent from every speaker, with a defined term
10Amazon PollyNo price published. Brand Voice is a sales-led custom engagementWeak as documented — and nothing found is Polly-specific
On this page

How we ranked, and what we do not claim

Three documented properties, weighted in this order: consent enforcement, whether the route is self-serve, and price.

Consent leads deliberately, and the reason is not squeamishness. A cloning product that verifies the speaker consented is a materially different thing from one that asks you to promise it did — different in legal exposure, different in what happens when a voice is misused, and different in what your own customers can do with it. Ranking on price alone would put the provider with no documented consent requirement near the top, which would be a bad recommendation dressed as an objective one.

What this ranking is not: it is not a quality ranking. Our standardized listening benchmark is designed but not yet running, so nothing here says which cloned voices sound most like their source. That is precisely the question a naturalness benchmark answers, and we have not run one. No numeric scores appear anywhere on this page.

Grouped by what the provider actually does, rather than what it says it values.

Speechify has the strongest technical enforcement we documented anywhere. Verbatim: “Consent is verified, not asserted. You never send a checkbox or a signed form: the speaker reads a phrase Speechify issues, and the recording of them doing so is checked and retained as the consent record for the voice.” The obvious workaround is closed explicitly: “Batch or unattended creation. There is no consent path without a speaker present.”

Azure Speech enforces a mandatory recorded statement for both of its cloning tiers, backed by biometric matching against the training audio. That is the most heavily enforced regime in this comparison — and it sits behind an eligibility gate that excludes anyone without a Microsoft account team.

OpenAI requires two separate recordings to create a custom voice, structurally enforcing the consent step rather than accepting an attestation.

Google Cloud requires a recorded, scripted consent statement submitted alongside the training audio.

ElevenLabs takes a different route to the same protection: professional cloning is restricted to your own voice, with identity verification. Verbatim: “You can only create a Professional Voice Clone of your own voice. Even with their consent, you cannot clone someone else’s voice.” That is narrower than a consent process — it forecloses third-party cloning at the professional tier entirely.

Murf AI requires explicit written consent from every speaker in the training audio, with a defined term, as part of an enterprise engagement rather than a self-serve flow.

Cartesia places the obligation on you contractually — its Acceptable Use Policy requires that you submit “your own voice and audio recordings or those of others with explicit consent” — while its Clone Voice API contains no occurrence of the word “consent” and no attestation field. The request takes a clip, a name, a language and metadata.

Resemble AI markets consent tooling, but its Terms are discretionary and set a verbal standard: “Resemble may require consent form the individual or third party whose voice is being cloned. Consent needs to be verbal, unless otherwise stated by Resemble.” The typographical error is the vendor’s own. Its product page describes something stricter; where a contract and a product page disagree, the contract governs.

Amazon Polly has no Polly-specific consent clause at all. The only consent language we found is a general prohibition in the AWS Responsible AI Policy against depicting “a person’s voice or likeness without their consent” — an unnumbered bullet covering the whole AI service family.

MiniMax gates cloning on the developer’s verified identity, which is a real abuse control on the account. But we found no voice-owner consent requirement published anywhere in its developer documentation, across two independent verification passes. The cloning flow takes an uploaded audio file and returns an identifier.

What this means for you: if you are cloning your own voice, this section may not change your decision. If you are cloning anyone else’s — a client, a colleague, a voice actor — the provider that verifies consent protects you as much as the speaker, because the consent record exists independently of your word for it.

What cloning actually costs

Documented routes to a cloned voice — verified September 1, 2026
ProviderEntry routeSelf-serve?Sample required
Cartesia$5/month ProYes10 seconds (instant); 30 minutes (professional)
ElevenLabs$6/month StarterYesInstant from Starter; professional from $22 Creator
Speechify$10/month API StarterYesSpeaker must record a phrase the vendor issues
MiniMaxPay-as-you-go, no subscriptionYesNot documented as a fixed requirement
Google Cloud$60 per 1M characters — allow-listedNoAudio plus a scripted recorded consent statement
Resemble AIContradictory: “first clone free, then $2 each” against a $1,000/month plan gateDisputedNot reconciled on any single page
Azure SpeechNot priceable without accessNoRecorded statement plus biometric match
OpenAINo price publishedNoTwo separate recordings
Murf AIEnterprise engagementNoStudio-quality recordings, multi-week turnaround
Amazon PollyNo price publishedNoNot published

Four of these ten publish no price for cloning at all. Amazon, Azure and OpenAI route the question to sales without a figure; Resemble publishes two figures that cannot be reconciled — a changelog dated May 8, 2026 saying “first clone free, then $2 each”, and documentation stating “The Voice Cloning API requires a Business plan or higher”, that plan being $1,000 a month with bullets that never mention voice. We publish both and label the conflict rather than picking one.

This page ranks providers on consent and licensing. What the process actually requires of you — how much audio each tier needs, how long it takes, the exact consent script you must record, and what happens to the clone afterwards — is set out separately in our guide to how AI voice cloning works.

Instant against professional cloning

Where a provider offers both, the split is consistent: a short sample for a fast approximate clone, or a long recording session for a higher-fidelity one.

Cartesia documents the clearest version: instant cloning from a 10-second sample on the $5 Pro plan, professional cloning from 30 minutes of audio on the $49 Startup plan, with professional clones capped at 2 on Startup and 4 on Scale.

ElevenLabs splits by plan rather than sample: instant from $6 Starter, professional from $22 Creator, with professional restricted to your own voice.

We cannot tell you whether professional cloning sounds meaningfully better than instant cloning at any of these providers. That comparison requires listening, and our benchmark has not run. What the documentation establishes is the input cost — seconds against half an hour of recording — and the price step.

Feature matrix

Documented cloning features — verified September 1, 2026
ProviderThird-party voicesClone persistsExport the cloneWatermarked output
ElevenLabsInstant only — professional is own-voiceYesNo — clones cannot be downloadedNot documented
SpeechifyYes, with verified consentYesConsumer app: no MP3 downloadNot documented
CartesiaYes, on your warrantyYes — survives model migrationNot documentedNot documented
MiniMaxYesNo — expires unless used in synthesisNot documentedMainland endpoint only
Resemble AIYesNot documented as expiringNot documentedYes, on open-source output
Google CloudYes, with recorded consentYesNot documentedNot documented
Azure SpeechYes, with biometric verificationYesNot documentedNot documented
OpenAIYes, two recordings requiredYesNot documentedAnti-evasion duty published
Murf AIYes, with written consentYesNot documentedNot documented
Amazon PollyBrand Voice engagementYesNot documentedNot documented

What cloning is good and bad at today

Good at:

Bad at:

Suitable and unsuitable use cases

Where cloning fits — editorial judgment from documentation, September 1, 2026
Use caseFitWhich provider
Cloning your own voice for your own contentStrongElevenLabs $6/month or Cartesia $5/month
A product where your users clone their voicesStrong on consent, check the termsSpeechify verifies by recording — but its API terms appear to forbid end-user uploads its own docs design for
Cloning a client’s or actor’s voice commerciallyMixedUse a provider that verifies consent; Murf’s written-consent regime suits contracted talent
Enterprise brand voiceStrong, at a price you must ask forAmazon Brand Voice, Azure custom voice, or Murf — none publishes a figure
Cheapest possible cloningWeak recommendationMiniMax is cheapest and documents no voice-owner consent requirement. We do not recommend it on price alone
Cloning without vendor lock-inWeak everywhereClones generally cannot be exported. Self-hosting an open-source model is the only route that avoids it

If none of these fits

Resemble AI publishes MIT-licensed open-source speech models with an express commercial grant, self-hostable with no account and no usage caps. That is the only route here that avoids both the vendor lock-in of an unexportable clone and any per-clone fee — at the cost of running the models yourself, and with the consent obligation resting entirely on you.

The provider-by-provider detail behind every claim on this page is in the individual reviews. The ten-provider view across all dimensions is the flagship ranking, and the free-tier question is covered separately in best free AI voice generators — no free tier here includes cloning.

Frequently asked questions

Did you test how these cloned voices sound?

No. We created no accounts and cloned no voices. Every statement comes from vendor documentation read on September 1, 2026. Our listening benchmark is designed but not running, so nothing here compares clone fidelity.

What is the cheapest way to clone a voice?

Cartesia at $5/month is the cheapest self-serve subscription route, with ElevenLabs at $6. MiniMax’s pay-as-you-go rapid cloning is cheaper still and needs no subscription — but we found no voice-owner consent requirement published in its documentation, and we do not recommend cloning on price alone.

Can I clone someone else’s voice?

Legally that depends on your jurisdiction and their consent, which is not a question this page can answer. Technically, ElevenLabs forecloses it at the professional tier — own voice only. Several providers require a recorded consent statement from the speaker before a voice can be created. Others take your warranty and check nothing.

Which provider enforces consent most strictly?

Azure Speech, by documentation: a mandatory recorded statement for both tiers backed by biometric matching. Speechify has the strongest enforcement reachable self-serve, since Azure’s cloning is closed to anyone without a Microsoft account team.

Can I download or move a cloned voice?

Generally no. ElevenLabs states it directly: “You cannot download or export your voice clones as standalone files.” Treat a cloned voice as tied to the vendor you made it with.

Is any cloning available on a free tier?

No. No free tier among the ten providers we cover includes voice cloning.

Why is Resemble AI ranked fifth when it markets cloning heavily?

Because its documented access route cannot be reconciled: a changelog says the first clone is free and further clones cost $2 each, while its documentation states the cloning API requires a $1,000-a-month Business plan whose published bullets never mention voice. Both are official and neither references the other.

What about instant versus professional cloning?

The documented difference is the input: seconds of audio against roughly half an hour, at a higher price. Whether the result sounds meaningfully better is a listening question, and our benchmark has not run.

Sources and what we could not verify

Every claim on this page was read from official vendor sources on September 1, 2026, in a verification pass covering all ten providers with an adversarial re-check of every price and access claim. The per-provider detail, with full quotations, is in the individual reviews linked throughout. How we verify anything is described on the methodology page.

What we could not verify

Honesty about gaps beats a page that looks complete. Genuinely unresolved as of September 1, 2026:

Change log

September 2, 2026 — First publication. All prices, access gates and consent requirements read from official vendor sources on September 1, 2026 and dated accordingly.

Published September 2, 2026 · This page shows no cloned-voice samples, because none exist: our benchmark has not run and we generate no audio. When it does, this page gains measured comparisons of clone fidelity and the ranking is revisited with a dated entry (methodology). Independence note: this page contains no affiliate links, and no vendor paid for placement or influenced the order — see how we make money and our editorial policy.