voice
Best AI Voice Clone
The best AI voice cloning tools as of June 2026 — ElevenLabs, Descript Overdub, Resemble AI, Cartesia, Fish Audio and open-source models like Chatterbox, with clone quality, sample requirements, pricing, and the consent and fraud risks you need to know.
Quick answer: For the highest cloning quality, ElevenLabs remains the benchmark — its Professional Voice Cloning produces near-indistinguishable results from 30+ minutes of audio. For podcasters and video editors, Descript Overdub builds cloning into a text-based editor and is now included on every plan. For enterprise security, Resemble AI leads with watermarking (Verify) and deepfake detection (Detect). For real-time voice agents, Cartesia Sonic 3.5 clones from a 3-second clip and runs at roughly 82ms. For the best value and open weights, Fish Audio’s OpenAudio S1 and Resemble’s Chatterbox match premium quality at a fraction of the cost. The strongest free, self-hosted option is Chatterbox or Fish Audio’s MIT-licensed weights.
Voice cloning is distinct from general text-to-speech: instead of picking a stock voice, you create a synthetic replica of one specific person’s voice — your own, a consenting actor’s, or, with permission, someone else’s. That carries technical, ethical and legal requirements this guide covers in full.
One caveat before you clone: a cloned voice in the wrong hands is dangerous. AI voice-cloning scams have surged — one in 10 Americans report having experienced a voice-clone scam (CNN, May 2026), and McAfee found that three seconds of audio is enough to produce an 85% voice match. Choose platforms with real consent verification, protect your voice samples, and agree a family “safe word” for emergency verification.
This guide covers the full stack — premium creator tools, enterprise security platforms, real-time APIs and open-source models you can self-host — with clone quality, sample requirements, current pricing and the safety picture. Two things have changed the landscape since our last update: PlayHT shut down permanently on 31 December 2025 after Meta acquihired its team, and the legal reckoning hardened — nine biometric-privacy class actions landed in May 2026, and the EU AI Act’s synthetic-audio labelling rule takes effect in August 2026.
The current state of AI voice cloning: June 2026
Voice cloning has crossed from novelty to commodity. What once needed hours of studio recording now works from three to thirty seconds of audio for an instant clone, and from 30 minutes for a near-perfect one. The AI voice-cloning market is estimated at roughly $4.06 billion in 2026, up from $3.28 billion in 2025, growing about 24% a year (The Business Research Company).
Five shifts define the current moment.
1. Instant cloning is “good enough” for most work. Few-shot models clone a usable voice from a single short clip with no training step. ElevenLabs Instant Voice Cloning needs 1–2 minutes, Cartesia and Qwen3-TTS clone from 3 seconds, and MiniMax from as little as 5. The quality gap to a 30-minute professional clone now only matters for premium, long-form or accent-heavy work.
2. PlayHT is gone. Meta acquihired the PlayAI team in July 2025 and permanently terminated the service on 31 December 2025, deleting all accounts, voice clones and audio with no migration path. The platform that topped many “largest voice library” lists a year ago no longer exists — if you see it recommended elsewhere, that guide is stale.
3. The quality crown is genuinely contested. ElevenLabs is no longer the automatic winner. Resemble AI’s open-source Chatterbox beat ElevenLabs in 63.75% of blind preference tests, and Fish Audio claims its OpenAudio S1 ranks first on the public TTS-Arena while costing roughly a sixth as much. On clean English, several tools are now effectively tied; price and workflow decide.
4. The legal reckoning intensified. Between 11 and 13 May 2026, nine class-action lawsuits were filed in Chicago against major voice-AI companies, including ElevenLabs, alleging voices were used for training without consent under Illinois’ Biometric Information Privacy Act. The EU AI Act’s Article 50, which mandates machine-readable labelling of synthetic audio, reaches full application in August 2026, and 46 US states now have deepfake laws.
5. Fraud scaled with the technology. Deepfake-enabled vishing rose more than 1,600% in early 2025, enterprises now lose an average of about $680,000 per voice-fraud attack, and a single deepfake video call cost engineering firm Arup $25.6 million. The same instant-cloning convenience that helps creators also powers “grandparent” and CEO-impersonation scams.
Top AI voice cloning tools (June 2026)
Clone quality is partly subjective, so the ratings below combine blind-preference testing (where it exists), sample requirements and current pricing. Treat vendor “indistinguishable” claims as a ceiling and blind-test arenas as the floor.
| Rank | Platform | Clone quality | Min. audio | Starting price | Best for |
|---|---|---|---|---|---|
| 1 | ElevenLabs | Excellent | 1 min instant / 30 min pro | $6/mo | Maximum quality, audiobooks |
| 2 | Descript Overdub | Very good | Brief Voice ID + audio | $16/mo | Podcasters, video editors |
| 3 | Resemble AI | Very good | 10–30 sec rapid | ~$20/mo + per clone | Enterprise, security, detection |
| 4 | Fish Audio (OpenAudio S1) | Very good | 10–30 sec | Free / $11/mo | Value, open weights, multilingual |
| 5 | Cartesia Sonic 3.5 | Very good | 3 sec | $5/mo | Real-time voice agents |
| 6 | MiniMax (speech-02) | Very good | 5–10 sec | Usage-based | Low-cost multilingual cloning |
| 7 | Speechify | Good | 20–30 sec | $29/mo | Consumer, accessibility |
| 8 | Respeecher | Excellent | 30 min–2 hr | from ~$15/mo | Film, TV, speech-to-speech |
Two things stand out. The open-weight options (Fish Audio, Chatterbox) now sit alongside the premium tools on quality, which did not exist a year ago. And the cheapest credible clone is no longer ElevenLabs — Fish Audio, MiniMax and self-hosted models undercut it heavily while staying close on output.
Voice cloning tools: detailed breakdowns
1. ElevenLabs — best overall quality
Price: Free (10K credits) | Starter $6 | Creator $22 | Pro $99 | Scale $299 | Business $990 Cloning: Instant (1–2 min) and Professional (30 min minimum, 3 hr optimal) Best for: Audiobooks, premium content, maximum fidelity
ElevenLabs still sets the cloning benchmark, and its Eleven v3 model (general availability 14 March 2026) adds inline audio tags such as [whispers] and [laughs] plus support for 70+ languages. The company raised a $500 million Series D at an $11 billion valuation in February 2026 and crossed $500 million in annual recurring revenue by May.
Two cloning tiers matter. Instant Voice Cloning uses few-shot adaptation: upload 1–2 minutes of clean audio and the clone is ready immediately, at roughly 85–90% fidelity — though it can struggle with very distinctive voices or strong accents. Professional Voice Cloning fine-tunes on 30 minutes to 3 hours and produces results most listeners cannot distinguish from the original.
Consent and limitations. Professional Voice Cloning requires a “Voice Captcha” verifying you own the voice; Instant cloning relies on checkbox self-attestation, which Consumer Reports called “no meaningful technical barrier.” The credit system remains the top complaint — failed generations consume credits, so real costs often exceed the advertised rate. ElevenLabs also faces a BIPA class action filed by seven journalists in May 2026.
Best for: Audiobook producers, YouTube creators and anyone where voice quality directly drives revenue.
2. Descript Overdub — best for podcasters and editors
Price: Free | Hobbyist $16 | Creator $24 | Business $50 (per month, annual) Cloning: Read a brief Voice ID statement, or upload existing audio Best for: Podcasters, video editors, content repurposing
Descript takes a different approach: cloning lives inside a full text-based audio and video editor, so you fix a flubbed line by retyping it and Overdub generates it in your voice, matched to the surrounding audio. Overdub is now included on every plan, and you can create a voice from existing project audio rather than reading a long script. Free and Creator accounts get a 1,000-word vocabulary; Pro accounts are unlimited.
Consumer Reports identified Descript as one of only two platforms (with Resemble AI) that had meaningful technical barriers against non-consensual cloning, thanks to its consent-script verification. Pure speech quality trails ElevenLabs slightly, and it is English-centric.
Best for: Podcasters fixing mistakes without re-recording, and creators already editing in Descript.
3. Resemble AI — best for enterprise security
Price: Flex usage ($0.0005/synthesis-second), voice clones ~$2–5/month each, team seats ~$20/user/month; Enterprise custom Cloning: Rapid (10–30 sec) and Professional Best for: Enterprise, regulated industries, anyone who needs detection and watermarking
Resemble AI competes on security, not just synthesis. It pairs voice cloning with two defensive products: Verify, an inaudible watermark embedded in every Resemble-generated clip, and Detect, a deepfake detector that claims 98.1% accuracy on the ASVspoof 2021 benchmark — with the multimodal DETECT-3B Omni reporting 98% across audio, video and images, trained against 160+ generative systems. Resemble also publishes the open-source Chatterbox model and offers streaming TTS at roughly 75ms via Chatterbox Turbo.
Best for: Enterprises worried about voice-theft liability, gaming studios and regulated industries that need watermarking and detection alongside cloning.
4. Fish Audio (OpenAudio S1) — best value and open weights
Price: Free (7 min) | Plus $11 (200 min) | Pro $75 | Max $749 (per month); yearly saves 33% Cloning: Zero- and few-shot from 10–30 seconds Best for: Cost-sensitive creators, developers, multilingual projects
Fish Audio is the breakout value play of 2026. Its OpenAudio S1 and S2 models clone a high-fidelity voice from 10–30 seconds of audio in under a minute, with an emotion-tag system and strong multilingual performance. Fish reports its model ranking first on the public TTS-Arena and beating ElevenLabs in blind A/B testing at roughly a sixth of the API cost — a vendor claim, but consistent with several independent 2026 round-ups. The weights are MIT-licensed and run on consumer GPUs with 8GB+ of VRAM, so you can self-host for free.
Best for: Creators and developers who want near-premium cloning without premium pricing, and anyone who needs open weights.
5. Cartesia Sonic 3.5 — best for real-time voice agents
Price: From $5/month, roughly $35 per million characters Cloning: Instant, from a 3-second clip Best for: Phone agents, live dialogue, developers bringing their own stack
Cartesia is the latency leader. Its Sonic 3.5 model, built on a state-space architecture, reaches about 82ms end-to-end, and Sonic Turbo pushes that to roughly 40ms — fast enough for natural turn-taking in a live call. Instant voice cloning works from a 3-second reference. Cartesia is built for developers rather than creators, so there is no timeline editor or stock-voice library; it competes on speed and API quality.
Best for: Real-time voice agents and customer-support lines where you supply your own speech-to-text and language model.
6. MiniMax (speech-02) — best low-cost multilingual cloning
Price: Usage-based, roughly half ElevenLabs at HD parity and a quarter at Turbo Cloning: Instant, from 5–10 seconds Best for: High-volume, multilingual, budget-conscious cloning
MiniMax’s speech-02 models (HD and Turbo) clone a voice from a 5–10 second clip and assign an instant voice ID for reuse, across 40+ languages, at roughly half the cost of ElevenLabs. Output quality is close to the premium tier for most content, which makes MiniMax a strong choice when you need many cloned voices or non-English coverage cheaply.
Best for: Developers and teams cloning at scale across languages.
7. Speechify — best for consumers and accessibility
Price: Premium $29/month (reader, includes cloning and OCR); Studio sold separately Cloning: 20–30 seconds Best for: Personal clones, accessibility, listening
Speechify won a 2025 Apple Design Award and leads the text-to-speech reader market; its SIMBA 3.0 model broke into the Artificial Analysis arena top 10. Speechify Premium ($29/month) bundles a personal voice clone with the reader and OCR; Speechify Studio is the separate product for commercial voice-overs — do not confuse the two.
Best for: Consumers wanting a simple personal clone and accessibility features.
8. Respeecher — best for film and TV
Price: From roughly $15/month, pay-as-you-go available Cloning: Speech-to-speech, 30 minutes to 2 hours of training Best for: Film, TV, games, high-production content
Respeecher is the Emmy-winning speech-to-speech technology behind Hollywood voice recreations — young Luke Skywalker in The Mandalorian and James Earl Jones’s Darth Vader in Obi-Wan Kenobi. Rather than text-to-speech, an actor performs the lines and Respeecher transforms their voice into the target, capturing performance nuance that pure TTS misses. The workflow suits professional pipelines, not quick content.
Best for: Studios and high-budget creative projects needing performance-driven cloning.
Note on a tool we no longer recommend: PlayHT (Play.ht), a top pick in our previous guide, was permanently shut down on 31 December 2025 after Meta acquihired its team. Do not start new projects on it.
Open-source and self-hosted voice cloning
For privacy, cost control or air-gapped work, open models now rival commercial tools. The open frontier moved well past the old Coqui-only days.
- Chatterbox (Resemble AI) — MIT-licensed, beat ElevenLabs in 63.75% of blind tests, clones from a 10-second clip and runs sub-200ms. English-first, with Chatterbox Multilingual covering 23 languages and Chatterbox Turbo fitting in roughly 6GB of VRAM. Every output is watermarked.
- Fish Audio OpenAudio S1 / S2 — MIT-licensed weights, 10–30 second cloning, emotion tags and strong multilingual output; runs on an 8GB+ GPU.
- Qwen3-TTS — Apache 2.0 licence with permissive cloning from 3-second samples.
- Coqui XTTS-v2 — the long-standing option, cloning from 6 seconds across 17 languages under an MIT licence; the company closed but the project is still maintained. Quality now trails Chatterbox and Fish Audio.
Note that some of the best open models, such as Kokoro-82M, deliberately omit voice cloning for safety — choose Chatterbox or Fish Audio if cloning is the point.
Big Tech voice cloning: still gated
The largest companies keep cloning behind approval gates, which is what created the opening for ElevenLabs, Resemble AI and others.
OpenAI’s Voice Engine, which can clone from 15 seconds, has stayed in limited preview with no public access since 2024. Google Cloud’s Instant Custom Voice (Chirp 3) clones from about 10 seconds across 30+ languages but requires a sales contact and a mandatory consent recording. Microsoft Azure Custom Neural Voice is generally available at roughly $24 per million characters but needs application approval and consent files. Amazon Polly Brand Voice is engagement-based, not self-service. Microsoft’s VALL-E 2 research reached “human parity” in 2024, but never shipped as a public product.
How voice cloning works
Understanding the inputs helps you get a better clone.
| Quality level | Audio required | Typical fidelity | Processing |
|---|---|---|---|
| Zero-shot / instant | 3–30 seconds | 70–85% | Seconds to minutes |
| Basic | 1–3 minutes | 85–90% | Minutes |
| Professional | 30–60 minutes | 95–99% | Hours to weeks |
| Studio-grade | 3+ hours | 99%+ | Days to weeks |
The sweet spot for most uses is a clean, varied 30-minute sample; beyond about three hours, gains are marginal. Quality depends more on audio cleanliness than quantity — a 30-second clean recording beats five noisy minutes. Record lossless (WAV/FLAC) at 44.1kHz or higher in a quiet room with a consistent microphone and distance, use natural and varied speech, and remove ums, false starts and background noise before training.
The ethics and safety of voice cloning
Your voice is a biometric identifier: once cloned, it cannot be reset like a password. The risks are no longer hypothetical.
The fraud landscape
- One in 10 Americans report having experienced a voice-clone scam (CNN, May 2026).
- Just three seconds of audio yields an 85% voice match, per McAfee research.
- Deepfake-enabled vishing rose more than 1,600% in early 2025, and enterprises lose an average of about $680,000 per voice-fraud attack.
- A single deepfake video call cost engineering firm Arup $25.6 million.
- US consumers reported losing more than $12.5 billion to fraud in 2024, up 25% year on year (FTC).
Platform verification is often weak
Consumer Reports tested six major platforms and found only two — Descript and Resemble AI — had meaningful technical barriers against cloning a voice without consent; the other four relied on a checkbox. ElevenLabs verifies ownership for Professional clones (Voice Captcha) but not for Instant clones. Resemble requires a matching consent clip; Descript requires a spoken consent script. Most others simply ask you to tick “I have the right to clone this voice.”
Regulation is hardening
The EU AI Act classifies voice cloning as high-risk and, under Article 50, mandates machine-readable labelling of synthetic audio from August 2026. In the US, the FCC has ruled AI voices in robocalls illegal under the TCPA — the fake-Biden primary robocall drew a $6 million fine — and the federal NO FAKES Act (reintroduced April 2025) would create a Digital Replica Right, though it has not passed. At state level, Tennessee’s ELVIS Act protects voice as a property right, California’s AB 2602 voids vague digital-replica contract terms, and 46 states have deepfake laws. The May 2026 BIPA class actions show the litigation risk is now live.
Protecting yourself
Agree a family safe word for emergency verification and never act on voice alone; if a call is urgent and asks for money, hang up and call back on a known number. Limit the voice recordings you post publicly. For organisations, prefer platforms with built-in watermarking (Resemble’s Verify, or detection via Detect), obtain explicit written consent with scope limits, and label all synthetic audio. No watermark is foolproof — a 2025 study defeated every scheme it tested across 109 configurations — so treat it as one layer, not a solution.
Feature comparison: the full matrix
| Platform | Clone type | Min. audio | Watermark/detection | Real-time | Open weights | Headline price |
|---|---|---|---|---|---|---|
| ElevenLabs | Instant + Professional | 1–30 min | No (audio-tag only) | Flash v2.5 | No | $6–990/mo |
| Descript | Overdub | Voice ID + audio | No | No | No | $16–50/mo |
| Resemble AI | Rapid + Pro | 10–30 sec | Yes (Verify + Detect) | ~75ms | Chatterbox only | ~$20/mo + per clone |
| Fish Audio | Zero/few-shot | 10–30 sec | No | Yes | Yes (MIT) | Free–$749/mo |
| Cartesia | Instant | 3 sec | No | ~82ms | No | $5/mo + usage |
| MiniMax | Instant | 5–10 sec | No | Turbo | No | Usage-based |
| Speechify | Instant | 20–30 sec | No | No | No | $29/mo |
| Chatterbox | Open-source | 10 sec | Yes (built-in) | Sub-200ms | Yes (MIT) | Free |
Use-case recommendations
For podcasters and video editors
Winner: Descript Overdub ($16/month)
Cloning inside a text-based editor lets you fix mistakes by retyping, with consent verification built in. Alternative: ElevenLabs Creator ($22/month) for higher pure-clone quality without the editing suite.
For audiobooks and premium content
Winner: ElevenLabs Professional Voice Cloning (Creator $22/month)
A 30-minute-plus clone meets ACX/Audible specifications and stays consistent across long-form narration. Alternative: Respeecher for performance-driven speech-to-speech. Note that listeners still prefer human narration for fiction; AI suits non-fiction and backlist titles.
For enterprise and security
Winner: Resemble AI
Watermarking (Verify), deepfake detection (Detect) and consent-clip verification reduce voice-theft liability. Alternative: WellSaid Labs (from ~$49/month), which offers only pre-cleared, paid-actor voices — no personal cloning means no consent problem.
For real-time voice agents
Winner: Cartesia Sonic 3.5
A 3-second clone at roughly 82ms suits live phone and support agents. Alternative: MiniMax speech-02 Turbo for cheaper multilingual real-time cloning.
For value and multilingual work
Winner: Fish Audio (OpenAudio S1) or MiniMax
Near-premium quality at a fraction of ElevenLabs’ price, with open weights (Fish) or 40+ languages cheaply (MiniMax). Alternative: self-hosted Chatterbox for zero per-use cost.
For film, TV and gaming
Winner: Respeecher
Emmy-winning speech-to-speech captures real performance for screen work. Alternative: Resemble AI for game audio with emotion control and watermarking.
For consumers and accessibility
Winner: Speechify Premium ($29/month)
A simple personal clone bundled with the best reader app and OCR. Alternative: ElevenLabs Starter ($6/month) for higher-quality instant clones.
For privacy and self-hosting
Winner: Chatterbox or Fish Audio (free, MIT)
Both run locally with no per-use cost and no audio leaving your machine; Chatterbox watermarks output, Fish Audio adds emotion tags and multilingual range.
Pricing comparison
| Platform | Free tier | Starting paid | Clone access | API / usage |
|---|---|---|---|---|
| ElevenLabs | 10K credits/mo | $6/mo | Instant from $6, Pro from $22 | Plan-bound credits |
| Descript | Yes | $16/mo | All plans (Overdub) | Enterprise |
| Resemble AI | Limited | ~$20/mo seat | Rapid + Pro | $0.0005/synthesis-sec |
| Fish Audio | 7 min | $11/mo | All plans | MIT self-host or cloud |
| Cartesia | Trial | $5/mo | Instant | ~$35/M chars |
| MiniMax | Trial | Usage-based | Instant | ~½ ElevenLabs cost |
| Speechify | Limited | $29/mo | Premium | $10/M chars |
| Respeecher | Pay-as-you-go | ~$15/mo | Speech-to-speech | Available |
Best free cloning: self-hosted Chatterbox or Fish Audio (MIT), or ElevenLabs Instant on Starter ($6/month). Best value at quality: Fish Audio OpenAudio S1 or MiniMax. Best enterprise value: Resemble AI, with watermarking and detection included.
What users actually think
ElevenLabs is the quality reference, but billing frustrates. Users praise the voices and cloning while complaining that retries on failed generations burn credits and real costs exceed advertised rates — “best voices, worst billing” is the recurring line.
Open weights changed the conversation. Chatterbox beating ElevenLabs in most blind tests, and Fish Audio undercutting it roughly sixfold, have made “you must pay premium for a good clone” obsolete for many creators.
Descript’s editing workflow keeps its fans. Podcasters describe Overdub plus text-based editing as a genuine time-saver, even if pure-clone quality is a notch below ElevenLabs.
Voice actors are anxious and litigious. The May 2026 BIPA suits reflect a profession that increasingly views unlicensed cloning as an existential threat, which is also pushing ethical, paid-actor options like WellSaid Labs.
Recent developments (2026)
PlayHT shut down (31 December 2025). Meta acquihired the team and terminated the service, deleting all clones and audio. A former category leader is simply gone.
ElevenLabs raised $500M at an $11B valuation (February 2026) and crossed $500M ARR, even as it faces a May 2026 BIPA class action over training data.
Open-weight cloning matured. Resemble’s Chatterbox (MIT) and Fish Audio’s OpenAudio S1/S2 (MIT) brought near-premium cloning to self-hosters, with Fish claiming a TTS-Arena top spot.
The legal calendar tightened. Nine BIPA class actions landed in Chicago in May 2026, and the EU AI Act’s synthetic-audio labelling rule takes effect in August 2026.
Frequently asked questions
How much audio do I need to clone a voice?
For instant or zero-shot cloning, 3–30 seconds produces usable results — Cartesia clones from 3 seconds, MiniMax from 5, and ElevenLabs Instant from 1–2 minutes. For professional quality that most listeners cannot distinguish from the original, you need 30 minutes minimum, with about 3 hours optimal. Clean audio matters more than quantity: a short clean recording beats a long noisy one.
Which AI voice clone is most accurate?
ElevenLabs Professional Voice Cloning remains the quality benchmark for near-indistinguishable clones from 30+ minutes of audio. But the field has tightened: Resemble AI’s open-source Chatterbox beat ElevenLabs in 63.75% of blind tests, and Fish Audio’s OpenAudio S1 claims the top TTS-Arena spot. On clean English the top tools are close, so workflow and price often decide.
Is AI voice cloning legal?
Cloning your own voice is legal. Cloning someone else’s requires their explicit consent and may still breach right-of-publicity, biometric-privacy or deepfake laws depending on your jurisdiction. Tennessee’s ELVIS Act, California’s AB 2602 and Illinois’ BIPA all create real exposure, and commercial use carries extra restrictions. Document consent thoroughly, and never clone a celebrity or public figure without rights clearance.
What is the best free AI voice cloning tool?
The strongest free options are self-hosted open-weight models — Resemble’s Chatterbox and Fish Audio’s OpenAudio S1, both MIT-licensed and able to run on a consumer GPU. Among hosted tools, ElevenLabs Instant cloning is available on the $6 Starter plan and Fish Audio offers a free tier. Free tiers usually prohibit commercial use, so check the terms.
How do voice-cloning scams work, and how do I protect myself?
Scammers harvest a few seconds of audio from social media or voicemail, clone the voice, and call relatives or colleagues claiming an emergency to demand urgent payment. One in 10 Americans report experiencing one. Protect yourself with a family safe word, by verifying urgent requests on a known number before acting, and by limiting public voice recordings. Treat any voice-only request for money or codes as suspect.
Can I clone a celebrity or someone else’s voice?
Only with their explicit, documented permission. Cloning a celebrity without rights clearance exposes you to liability under laws like Tennessee’s ELVIS Act, and reputable platforms should block it. The nine BIPA lawsuits filed in May 2026 show that even training on voices without consent is now being litigated.
Does voice cloning carry a watermark?
Some platforms add one. Resemble AI watermarks every output via Verify and can detect it with Detect, and open-source Chatterbox watermarks all output. Most others, including ElevenLabs Instant clones, do not embed a robust audio watermark. No watermark is fully tamper-proof — a 2025 study defeated every scheme it tested — so watermarking is a layer of protection, not a guarantee.
What happened to Play.ht (PlayHT)?
PlayHT was permanently shut down on 31 December 2025 after Meta acquihired the PlayAI team in July 2025, with all accounts, voice clones and audio deleted and no migration path. If another guide still recommends it, that guide is out of date. Use ElevenLabs, Fish Audio, Resemble AI or Cartesia instead.
Conclusion: how to choose in June 2026
Voice cloning has reached functional maturity, and the decision now turns on workflow, security and price rather than raw capability.
- Maximum quality: ElevenLabs Professional Voice Cloning ($22/month).
- Podcast and video editing: Descript Overdub ($16/month).
- Enterprise and security: Resemble AI, with watermarking and detection.
- Value and open weights: Fish Audio (OpenAudio S1) or self-hosted Chatterbox.
- Real-time agents: Cartesia Sonic 3.5 (3-second clone, ~82ms).
- Consumer and accessibility: Speechify Premium ($29/month).
- Film and TV: Respeecher speech-to-speech.
The non-negotiable is consent. Clone your own voice freely, get explicit permission for anyone else’s, protect your samples, and verify urgent voice requests independently — because the same three seconds of audio that makes a convenient clone also makes a convincing scam.
For related guides, see best AI voice generators, best AI for transcription, best AI music generators and the best AI apps.
This guide is updated as voice-cloning models launch and regulations change. Clone quality varies by source audio and use case; vendor “indistinguishable” claims run ahead of blind testing, and we cite sources inline. Pricing and availability are subject to change.