AI voice cloning is the process of training a speech model on recordings of a specific human voice so that it can generate new speech in that voice from text.
Quick answer: ElevenLabs offers Instant cloning on eligible paid plans, with own-voice-only Professional cloning from Creator ($22/month). Cartesia Pro starts at $5/month for Instant cloning. Kits Starter ($10/month) lists Professional cloning on its plan card, while its FAQ differs; confirm checkout entitlements. SpeechifyAI documents spoken speaker verification, which does not replace a complete commercial authorisation agreement.
Evaluation methodology
This guide examines cloning services alongside related speech-generation products. Not all offer self-service cloning. Six reviewed services document cloning access: ElevenLabs, Cartesia, SpeechifyAI, Fish Audio, Kits AI and Murf (Enterprise API). Deepgram, Voicemod and WellSaid are related alternatives. Resemble generation is discontinued; LOVO and Play.ht availability is unverified. No equivalent twelve-product hands-on ranking is claimed.
Pricing references are USD as of October 11, 2026. Billing units, cloning entitlements, commercial licences and consent requirements differ. Replica and Uberduck are outside this selected comparison; Suno is music generation and is not evaluated as a direct cloning service.
Voice cloning services at a glance
| Platform | Cloning workflow | Published starting price | Unit and restrictions |
|---|---|---|---|
| ElevenLabs | Instant and Professional; Professional is own-voice only | Starter $6/month; Creator $22/month | Credits; Professional from Creator; Free lacks commercial licence |
| Cartesia | Instant from Pro; Professional from Startup | Pro $5/month; Startup $49; Scale $299 | Credits; Startup 2 Professional slots, Scale 4 |
| SpeechifyAI | API cloning with spoken verification | Starter $10/month | 1.9M characters included; $10/additional million; separate consumer app |
| Fish Audio | Consumer cloning and developer API | Consumer Free; S2.1 Pro API $15/million UTF-8 bytes | Consumer credits and API bytes are separate |
| Kits AI | Professional listed on Starter plan card; Instant/Professional on Producer | Starter $10/month; Producer $30 | 15/60 download minutes; confirm conflicting FAQ entitlements |
| Murf AI | Enterprise API Instant cloning, workspace-enabled | Enterprise quote for cloning | General API synthesis prices do not buy cloning access |
Stock TTS, real-time voice effects and unavailable services are listed separately below.
Voice cloning services, reviewed
ElevenLabs
ElevenLabs offers Instant and Professional Voice Cloning with different workflows and safeguards. Professional requires identity verification, is limited to the user’s own voice and is available from Creator upward. Instant cloning has separate technical safeguards and usage rules.
ElevenLabs restricts Professional clones to the user’s own verified voice even when another person has consented. A speaker can create and verify a Professional clone on their account and share it through the supported private-sharing route. This rule should not be generalised to every cloning method.
The two cloning tiers differ in cost and turnaround. Instant Voice Cloning clones from less than two minutes of audio. Professional Voice Cloning trains a dedicated model on a large voice dataset, and ElevenLabs says fine-tuning “usually” takes 3 to 6 hours.
Pricing is credit-based, with credits shared across products and roughly one credit per character:
- Free - $0, 10,000 credits
- Starter - $6/month, 30,000 credits
- Creator - $22/month, 121,000 credits ($11 first month)
- Pro - $99/month, 600,000 credits
- Scale - $299/month, 1,800,000 credits, 3 seats
- Business - $990/month, 6,000,000 credits, 10 seats
- Enterprise - custom
Annual billing runs at roughly ten months’ price for twelve.
Credit consumption depends on the model and workflow. Estimate actual synthesis usage and approved commercial outputs before choosing a plan.
ElevenLabs documents model-dependent coverage: PVC supports all languages in the Eleven v4 model family - 90+ languages.
On commercial rights, read the rules before you publish. The free plan does not include a commercial licence; free-plan output must attribute ElevenLabs by including “elevenlabs.io” or “11.ai” in the title. All paid plans include a commercial licence, provided you are not using Beta Services - and Beta output cannot be used commercially at all.
ElevenLabs also runs a “no-go voices” safeguard designed to detect and prevent creation of specific voice clones, and states it is expanding that safeguard to other languages.
The limitation: credits make cost forecasting harder than a flat subscription, and the platform is organised around cloning and stock voices rather than treated studio recording. If you need a producer in a room, this is not that.
Cartesia
Cartesia provides API-first speech generation and cloning. Its current pricing features Sonic 3.6; advertised latency figures are model-level vendor measurements, not complete application latency.
The documentation is unusually candid. An Instant Voice Clone (IVC) requires 10 seconds of audio, up to 60 seconds if you want the speaker’s accent better retained, and needs the Pro plan or above. A Pro Voice Clone (PVC) requires 30 minutes or more of studio-quality single-speaker audio, with training taking up to 3 hours, and needs the Startup plan or above.
The constraints matter for game and animation work. A PVC is trained on one language at a time, so multilingual voice work means a separate recorded dataset per language. Cartesia advises varying emotion across the dataset so the clone is not locked to one register, and recommends recording raw with effects off, because added noise reduction and EQ can create unnatural artefacts. It also notes the clone picks up the mood of your clip - a cheerful take gives you a cheerful-sounding voice in everything it says.
Published prices are Pro $5/month, Startup $49 and Scale $299. Instant cloning starts on Pro; Professional cloning starts on Startup. The slots are tied to plan: Startup provides 2 PVC slots, Scale provides 4, Enterprise is custom, and Free and Pro have none. Existing clones keep working after a downgrade, but you cannot train a new one until you are back under the limit.
The limitation: that 30 minutes of studio audio per language is a real production commitment, and serious cloning starts at Startup rather than the Pro tier. Cartesia describes Sonic 3.6 as streaming “the first byte of audio in about 90ms.”
Speechify
Speechify is a reading and narration platform whose API enforces a spoken consent challenge and retains the consent recording as a durable record.
Cloning runs through a consent challenge: you POST the speaker’s full_name, they record themselves reading a returned phrase, and that recording plus the challenge ID is submitted with the sample. Speechify states it “is checked against the phrase and speaker, then retained as the consent record.”
The spoken record supports speaker verification and consent evidence, but does not establish every contractual permission, usage scope or statutory requirement. Retain the authorised purpose, term, distribution rights and revocation agreement separately.
The mechanics: a 10-30 second clean sample, instant self-serve cloning with no fine-tuning run, zero-shot and fine-tuned tiers, and a fine-tune option on hours of audio for signature narrators and brand voices. Cloning is included on paid plans from Starter upward with no separate per-clone charge - a free-tier attempt returns a 402 with the code voice_cloning_not_included.
Two details in the consent documentation tell you how seriously this is built. The consent recording must be the same person as the voice sample - a different voice reading the phrase is refused with consent_speaker_mismatch - and Speechify’s position on the old approach is blunt: the previous flow took a consent field carrying the speaker’s name and email, “asserted by the caller and checked by nothing.” That flow was switched off for every workspace on September 23, 2026, deliberately ahead of the standard 12-month sunset, because “an endpoint that clones a voice without checking the speaker agreed is a safety liability.”
One operational catch before you build on this: Speechify geo-restricts cloning by the IP address of the calling machine and returns 403 voice_cloning_unavailable_in_region where cloning is unavailable. Their docs are specific that when your backend calls the API, the location that counts is your server’s, not your end user’s.
The limitation: Six self-serve languages on Simba 3.0; Simba 3.2 is English only, with broader coverage negotiated through sales. A real gap if your plan is multilingual output from one recording.
SpeechifyAI’s developer plans are separate from its consumer reading app. Starter is $10/month with 1.9 million characters and $10 per additional million; Pro is $99/month with $8/million overage; Scale is $499/month with $6/million overage. Paid API plans include cloning. The $6 rate is not the entry-tier rate.
Fish Audio
Fish Audio is a low-cost voice platform that includes enhanced voice cloning on its paid tiers, billed in credits rather than characters or minutes.
Fish Audio separates consumer credits from API usage. Its developer documentation lists S2.1 Pro at $15 per million UTF-8 input bytes. English bytes and characters may be similar; non-ASCII characters can require multiple bytes. Language and encoding affect effective cost. Its commercial open-weight licence and enterprise self-hosting offering start from $10,000/month; consumer allowances are a separate purchase.
That open-weight angle is genuinely unusual here. Vendors such as Deepgram and Cartesia sell on-premises deployment of a proprietary model; Fish Audio sells you the weights.
One more number before you build against it: Fish Audio meters concurrency, at 5 concurrent requests below $100 of spend, 15 at $100, and 50 at $1,000. Their own planning table shows why that matters - an audiobook call holding a slot for 60 seconds yields roughly 5 requests per minute at the starter tier, against about 150 for a two-second companion reply. Long-form and interactive workloads cannot share a budget.
For a trial, confirm the free allowance and its restrictions. Do not assume a free developer offer means unlimited production API usage.
Fish Audio’s terms permit specified uses of content for model development. Confirm privacy, retention, training restrictions and deletion contractually before uploading a client’s master recording.
Regulated deployments require separate contractual, privacy and security review. Keep consumer generation minutes and developer UTF-8 byte billing distinct.
Kits AI
Kits AI is a music-and-voice platform whose paid tiers unlock instant and professional voice cloning alongside stem separation and vocal repair tools.
Its pricing is unusually clear:
- Starter - $10/month, 2 voice slots, 15 download minutes; plan card lists Professional Voice Cloning
- Producer - $30/month, unlimited voice slots, 60 download minutes, instant plus professional voice cloning
- Professional - $60/month, unlimited voice slots, unlimited download minutes
The free plan does not include Professional or Instant cloning. Starter’s $10 plan card lists Professional cloning, while the same page’s FAQ describes Instant cloning differently. Producer ($30/month) documents both methods with 60 download minutes and unlimited voice slots, subject to fair use. Confirm the entitlement at checkout rather than stating cloning only begins at $30.
A cancellation detail worth flagging: per Kits AI’s terms, when you cancel you keep added features until the end of the billing period and your custom AI voice clones will be locked. Export finished audio before cancelling.
The limitation: billing is in download minutes, not characters or generation credits, and long-form narration burns those faster than people expect. Note a discrepancy on Kits AI’s own site: the pricing page lists Starter at $10/month, while the voice-cloning page says paid plans start at $14.99/month. Get the real figure from checkout.
Murf AI
Murf offers Instant Voice Cloning through its Enterprise API. Its September 8, 2026 update documents a recording of up to 30 seconds, asynchronous creation and a reusable voice ID for Falcon 2 synthesis. Access is enabled per workspace and is not included in ordinary self-service pricing. See API documentation.
Murf also provides stock TTS, voice-changing and translation products. General synthesis rates and trial credits should not be treated as the purchase price for Enterprise cloning.
Alternatives to cloning a voice
Deepgram - related TTS alternative
Deepgram provides programmatic speech generation through Aura and Flux TTS. Its public pricing does not establish an equivalent self-service workflow for cloning a customer’s voice. Evaluate its character-based prices as TTS infrastructure, not a directly interchangeable cloning subscription.
Voicemod - real-time voice transformation
Voicelab provides real-time voice modification and custom effects. This differs from uploading a speaker sample and generating arbitrary text in a replicated voice. Confirm subscription entitlements for the intended live application.
WellSaid Labs - licensed stock voices
WellSaid offers licensed stock narration rather than a verified self-service cloning workflow. Published entry subscriptions start at $10/month billed annually, with Pro at $33/user/month annually and plan-specific download allowances. Licensed stock audio still requires adherence to its licence and accurate representation.
Discontinued and unverified services
Resemble AI - generation discontinued
Resemble’s official discontinued-product notice states that commercial voice generation ended and its focus shifted to detection, watermarking and multimodal security. It is not included in the current cloning-service table. Verify any migration or retention arrangements directly with the vendor.
LOVO and Play.ht - availability unverified
Their current commercial availability and subscription terms could not be independently established from the reviewed official URLs. Website access errors do not prove either company has permanently ceased operating; confirm availability before recommending or budgeting.
Consent verification and commercial authorisation
ElevenLabs documents own-voice identity verification for Professional cloning. SpeechifyAI documents a recorded spoken challenge and retirement of its unverified flow on September 23, 2026. These are technical controls, not proof of every legal or contractual permission. Other providers’ safeguards require workflow-specific confirmation.
Buyers should request written terms covering speaker authorisation, identity checks, permitted use, languages, distribution, term, training, retention, deletion, revocation and sharing. Confirm how controls apply to Instant versus Professional cloning and how the platform handles impersonation reports.
Likeness, voice and the law in 2026
Rights vary by jurisdiction, intended use and statutory exceptions. Written, informed permission is the recommended commercial standard when cloning another person’s voice, alongside review of privacy, publicity rights, contracts and platform restrictions. Licensed cloning is possible on some services even though ElevenLabs restricts its Professional workflow.
New York Civil Rights Law § 50 covers specified unauthorised advertising or trade uses of a living person’s name, portrait, picture, likeness or voice without written consent, with misdemeanor liability. Section 51 supplies civil remedies including injunctions, damages and potentially exemplary damages for knowing unlawful use; statutory exceptions matter.
Tennessee’s ELVIS Act recognises transferable voice and likeness property rights that survive death. Section 47-25-1104 provides initial ten-year post-death exclusivity, with continued protection through commercial exploitation subject to non-use termination rules. Ten years is not an unconditional expiration. Consult the official legislative text and current codification for §§ 47-25-1103–1104.
Under EU AI Act Article 50, relevant providers have machine-readable marking duties and deployers have separate disclosure duties for qualifying deepfakes. Role, content, technical exceptions and artistic-work provisions affect scope. The general application date is August 2, 2026; specified pre-existing systems have a limited marking transition to December 2, 2026. Potential fines of up to €15 million or 3% of worldwide annual turnover are statutory maxima subject to applicable rules and proportionality, not automatic penalties for every omitted disclosure.
Advertising rules also apply independently of voice ownership. The FTC Endorsement Guides address misleading endorsements; a synthetic testimonial presented as a genuine customer opinion can create deception concerns even when a voice licence exists.
Provenance and watermarking
Some providers use watermarks or origin verification, but implementation, robustness and detectability vary. ElevenLabs describes its own watermark and detector; this is not evidence that every service or third-party model uses equivalent controls.
C2PA specifies signed provenance for audio and other media. Supporting the standard does not mean every file automatically includes or retains Content Credentials after editing, transcoding or re-recording. Preserve consent records and original assets, and verify the actual delivery workflow.
When cloning is inappropriate
Avoid deployments without documented authorisation or a clear understanding of permitted use. Deceased speakers’ rights require jurisdiction-specific review; estate permission is not a universal substitute for checking rights and exceptions. Do not present synthetic endorsements or impersonations in misleading ways.
If speaker verification, licence scope or privacy requirements cannot be satisfied, consider licensed stock voices or conventional recording. Consent and provenance controls reduce particular risks but do not guarantee lawful use.
Choosing a workflow
Select a documented cloning workflow, confirm its subscription entitlement and retain the agreed usage scope. Stock TTS and live voice effects are alternatives with different capabilities. Evaluate commercial rights, privacy, verification and disclosure for the specific project rather than assuming the most expensive plan resolves every legal issue.
Sources
- Speechify Build Docs - Voice Cloning Overview (two-leg consent system, single-use challenge, sample requirements, geo-restriction by calling IP) (accessed October 10, 2026) - https://docs.speechify.ai/build/guides/voice-cloning/overview
- Speechify Build Docs - Voice Cloning Consent (challenge lifecycle, consent_speaker_mismatch, unverified flow switched off) (accessed October 10, 2026) - https://docs.speechify.ai/build/guides/voice-cloning/consent
- Speechify Build Docs - Voice Cloning API (required consent fields, versioning from 2026-09-13, model support) (accessed October 10, 2026) - https://docs.speechify.ai/build/voice-cloning-api
- Speechify - Voice Cloning product page (10-30s sample, 402 voice_cloning_not_included, 6 self-serve languages on Simba 3.0) (accessed October 10, 2026) - https://www.speechify.ai/voice-cloning/
- Speechify - Pricing (paid API tiers and tier-specific overage prices; voice cloning on paid plans) (accessed October 10, 2026) - https://www.speechify.ai/pricing/
- Fish Audio Docs - Models, Pricing and Rate Limits (s2.1-pro at $15.00 per M UTF-8 bytes; concurrency tiers of 5, 15 and 50 requests by spend threshold; voice design at $0.01 per successful request) (accessed October 10, 2026) - https://docs.fish.audio/developer-guide/models-pricing/pricing-and-rate-limits
- Deepgram - Flux TTS product page ($0.045 per 1,000 characters, 80ms latency, on-premises and private cloud deployment) (accessed October 10, 2026) - https://deepgram.com/product/text-to-speech
- Deepgram - Pricing (Aura-1 and Aura-2 per 1,000 characters; voice changer per minute; translation per 1,000 characters) (accessed October 10, 2026) - https://deepgram.com/pricing
- Voicemod - Voicelab Voice Maker (140+ voice effects, publish and remix community, real-time) (accessed October 10, 2026) - https://www.voicemod.net/en/voicelab/
- Voicemod - Pricing (subscription tiers) (accessed October 10, 2026) - https://www.voicemod.net/pricing/
- AI Act Explorer - Article 50: Transparency obligations for providers and deployers of certain AI systems (accessed October 10, 2026) - https://artificialintelligenceact.eu/article/50/
- AI Act Explorer - Article 113: Entry into Force and Application (Regulation applies from 2 August 2026) (accessed October 10, 2026) - https://artificialintelligenceact.eu/article/113/
- AI Act Explorer - Practical guide to the Article 50 transparency rules, including the AI Omnibus 2 December 2026 marking grace period and the artistic-work carve-out (accessed October 10, 2026) - https://artificialintelligenceact.eu/transparency-rules-article-50/
- European Commission, DG CONNECT - Code of Practice on Transparency of AI-generated Content (final code published 10 June 2026; provider and deployer rules; about 190 signatories by end of July 2026; EU labelling icons) (last updated 31 July 2026) - https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content
- eCFR - 16 CFR Part 255, Guides Concerning Use of Endorsements and Testimonials in Advertising, 88 FR 48102, July 26, 2023 (accessed October 10, 2026) - https://www.ecfr.gov/current/title-16/chapter-I/subchapter-B/part-255
- RightOfPublicity.com (Jonathan Faber) - Tennessee Code Annotated 47-25-1101 to 1108, Ensuring Likeness, Voice, and Image Security Act of 2024 (ELVIS Act), current through end of 2025 Reg. Sess. (accessed October 10, 2026) - https://rightofpublicity.com/statutes/tennessee
- RightOfPublicity.com - New York Civil Rights Law 50 and 51, right of privacy and action for injunction and damages, expressly covering voice (accessed October 10, 2026) - https://rightofpublicity.com/statutes/new-york
- RightOfPublicity.com - California Civil Code 3344.1, Astaire Celebrity Image Protection Act (accessed October 10, 2026) - https://rightofpublicity.com/statutes/california
- RightOfPublicity.com - Washington RCW 63.60.010 and 63.60.040, Personality Rights (voice as a property right; 10 years for individuals, 75 years for personalities) (accessed October 10, 2026) - https://rightofpublicity.com/statutes/washington
- Cartesia Docs - Voice Cloning documentation home (Sonic 3.6 first-byte latency of about 90ms, Ink 2 streaming STT, Managed Agents) (accessed October 10, 2026) - https://docs.cartesia.ai/
- California Legislative Information - AB 1836, Digital Replicas of Deceased Persons, full bill text (accessed October 10, 2026) - https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202320240AB1836
- C2PA - Content Credentials specification, covering audio alongside image and video (accessed October 10, 2026) - https://c2pa.org/specifications/specifications/2.2/specs/C2PA_Specification.html
- ElevenLabs - Pricing (plan tiers and credit allowances; annual billing) (accessed October 10, 2026) - https://elevenlabs.io/pricing
- ElevenLabs Help Center - Can I create a Professional Voice Clone of someone else’s voice? (own-voice-only verification rule) (accessed October 10, 2026) - https://elevenlabs.io/docs/help-center/product/voices/voice-cloning/can-i-create-a-professional-voice-clone-of-someone-elses-voice
- ElevenLabs Help Center - Can I publish the content I generate on the platform? (free plan has no commercial licence; attribution requirement; Beta restriction) (accessed October 10, 2026) - https://elevenlabs.io/docs/help-center/legal/can-i-publish-the-content-i-generate-on-the-platform
- ElevenLabs Help Center - What is watermarking, and why is ElevenLabs using it? (accessed October 10, 2026) - https://elevenlabs.io/docs/help-center/legal/audio-detector/what-is-watermarking-and-why-is-eleven-labs-using-it
- Cartesia - Pricing (accessed October 10, 2026) - https://cartesia.ai/pricing
- Cartesia Docs - Pro Voice Clone (30 minutes of audio, up to 3-hour training, Startup plan, PVC slot limits, one language per clone) (accessed October 10, 2026) - https://docs.cartesia.ai/build-with-cartesia/capability-guides/clone-voices-pro
- Cartesia Docs - Instant Voice Clone (10 seconds of audio, Pro plan or above, recording guidance) (accessed October 10, 2026) - https://docs.cartesia.ai/build-with-cartesia/capability-guides/clone-voices
- Murf AI - Pricing (Falcon 2, Gen-2, voice changer and translation rates per 1,000 characters) (accessed October 10, 2026) - https://murf.ai/pricing
- Murf AI - AI Voice Changer (200+ voices, 20+ languages, free tier of 10 minutes generation and transcription) (accessed October 10, 2026) - https://murf.ai/voice-changer
- Kits AI - Pricing (Starter $10, Producer $30, Professional $60; Starter card lists Professional cloning; FAQ entitlements differ; clone locking on cancellation) (accessed October 10, 2026) - https://www.kits.ai/pricing
- Kits AI - Voice Cloning (dataset preparation and training workflow; voice-cloning page states paid plans start at $14.99/month, higher than the Starter tier listed on the pricing page) (accessed October 10, 2026) - https://www.kits.ai/voice-cloning
- Fish Audio - Plan and pricing (credit tiers by generation minutes; enhanced voice cloning) (accessed October 10, 2026) - https://fish.audio/plan/
- Fish Audio - Developer platform ($15 per 1M UTF-8 bytes, S2.1 Pro API; verify current trial allowances, 30-second instant clone, open-weight self-hosting at $10,000/month with a $120,000-$150,000 effective annual floor) (accessed October 10, 2026) - https://fish.audio/developers
- Fish Audio - Terms of Service (content may be used to train or enhance Fish.Audio models; prohibition on representing content as entirely human generated) (accessed October 10, 2026) - https://fish.audio/terms
- WellSaid Labs - Pricing (stock voice library; no self-serve cloning) (accessed October 10, 2026) - https://wellsaidlabs.com/pricing
- Resemble AI - Why a former Voice AI company went all-in on deepfake detection (generation discontinued) (accessed October 10, 2026) - https://www.resemble.ai/resources/why-a-former-voice-ai-company-went-all-in-on-deepfake-detection
- Resemble AI - Voice cloning product page (generation discontinued) (accessed October 10, 2026) - https://www.resemble.ai/voice-cloning/
- Resemble AI - Pricing (detection-led product surface) (accessed October 10, 2026) - https://www.resemble.ai/pricing/
- Additional primary reference (revision October 11, 2026) - https://www.cartesia.ai/pricing
- Additional primary reference (revision October 11, 2026) - https://speechify.ai/pricing
- Additional primary reference (revision October 11, 2026) - https://murf.ai/api/docs/voices-styles/instant-voice-cloning
- Additional primary reference (revision October 11, 2026) - https://www.resemble.ai/product-notice
- Additional primary reference (revision October 11, 2026) - https://www.nysenate.gov/legislation/laws/CVR/50
- Additional primary reference (revision October 11, 2026) - https://www.nysenate.gov/legislation/laws/CVR/51
- Additional primary reference (revision October 11, 2026) - https://digital-strategy.ec.europa.eu/en/factpages/quick-facts-transparency-rules-ai-systems
- Additional primary reference (revision October 11, 2026) - https://capitol.tn.gov/bills/113/House/Journals/03042024RD44.pdf
