Best AI Voice Cloning Tools (2026): Prices, Free Tiers and Consent Rules, Compared

Short answer: to clone your own voice for narration or videos, ElevenLabs is the most complete option: instant cloning from its $6 Starter plan and a trained Professional Voice Clone from Creator at $22 a month. Fish Audio is the cheapest way to try cloning for free, though free output is for personal, non-commercial use only. Developers building a voice app should look at Cartesia, whose $5 Pro plan includes instant cloning and a commercial license. Podcasters who edit in a timeline get cloning inside Descript. Updated November 9, 2026.

How we checked: prices, credits, clone limits and consent rules come from each vendor's own pricing page, documentation, terms or help center, read November 9, 2026. Where a vendor page does not state something, we write "Not stated". We did not test these tools hands-on, so this guide compares published prices, limits and rules, not how natural each clone sounds. Prices change often; check the linked page before you buy.

Most vendors sell two levels, and the names differ:

• Instant clone. Built in seconds from a short sample. Cartesia's docs say 10 seconds is enough to start and up to 60 seconds helps keep the accent. ElevenLabs recommends 1 to 2 minutes of clean audio. Good for drafts, social clips and app prototypes.
• Professional (trained) clone. A model fine-tuned on much more audio. ElevenLabs asks for at least an hour, ideally close to three; Cartesia's Pro clone trains on 30 minutes or more. Worth it for audiobooks, courses and a voice you will use for years.

Record in a quiet room with one microphone and no music. Every vendor that gives guidance says the same thing: noise and reverb in the sample end up in the clone.

ElevenLabs (Stork listing) sells both clone types on one account. Its pricing page lists Free at 10,000 credits a month with no cloning and no commercial license; Starter at $6 a month ($1 for the first month, $5 a month on annual billing) with Instant Voice Cloning and a commercial license; Creator at $22 a month ($11 for the first month, $18.33 annual) with 121,000 credits and Professional Voice Cloning; and Pro at $99 with 600,000 credits.

For an instant clone, the docs recommend 1 to 2 minutes of clear audio, at least 10 seconds in total, and warn that going much past 3 minutes adds little. A Professional Voice Clone needs at least an hour of training audio and a spoken verification step to prove the voice is yours. Our ElevenLabs voice cloning guide walks through both.

Not for: cloning a colleague's or client's voice into your account. ElevenLabs says a Professional clone can only be of your own voice, "even with their consent". The owner has to make and verify the clone, then share it with you.

Fish Audio (Stork listing) is the only service here that includes cloning on its free plan. Its plans page lists Free at $0 with 8,000 credits a month (up to 7 minutes of generation, 500 characters per generation) and "Enhanced voice cloning". Plus is $132 billed annually ($11 a month) with 250,000 credits, up to 200 minutes of generation and 1 professional voice slot. Pro is $900 billed annually ($75 a month) with 2,000,000 credits, 3 seats and 5 professional voice slots.

Its developer docs ask for at least 10 seconds per clip and say a minute or two of clear, single-speaker speech improves fidelity. Fish Audio also publishes its model code, covered in the open-source section below.

Not for: monetized content on the free plan. The plans page lists "Commercial use" on Free, but Fish Audio's Terms limit free users to "your own internal, personal, non-commercial use". Pay before you publish anything that earns money.

Cartesia (Stork listing) sells a speech API built for voice agents and live apps. Its pricing page lists Free at 20K credits a month with no cloning; Pro at $5 a month with 100K credits, instant voice cloning and a "Commercial use license"; Startup at $49 with 1.25M credits and 2 Pro voice cloning slots; and Scale at $299 with 8M credits and 4 slots.

The cloning docs say 10 seconds of audio is enough to start, up to 60 seconds on Sonic 3.6 and newer, and 30 minutes or more for a Pro clone. Each clone is built in one language; you add other languages afterwards.

Not for: someone who wants an editor with a timeline and a Download button. Cartesia has a playground, but it is mostly an API for developers.

Descript (Stork listing) puts a voice clone inside a text-based audio and video editor, so you can fix a misspoken word by typing. Its pricing page lists a Free plan with limited text-to-speech and custom voice clones, 100 one-time AI credits and watermarked 720p export; Hobbyist at $16 a person a month billed annually or $24 monthly, which includes "AI Speech with custom voice clones"; and Creator at $24 annual or $35 monthly, with 800 AI credits a month.

To make a clone, Descript's help center says you record or upload a consent statement, which must be in English even if you will generate speech in another language. You cannot add more training audio to an existing clone; a new clone needs a new consent recording. How much training audio it uses is not stated.

Not for: developers who need an API, or anyone who only wants voice output and no editor.

The developer side of Speechify, listed on Stork, sells a text-to-speech API. Its API pricing lists Free at $0 with 500K characters a month and no cloning; Starter at $10 a month with 1.9M characters and cloning; Pro at $99 with 13.5M characters; and Scale at $499. Plans are monthly with no annual commitment.

What sets it apart is its consent flow. Speechify issues a single-use phrase, the speaker reads it aloud, and Speechify checks both the words and that the speaker matches the voice sample. In its words, "the person consenting has to be the person being cloned." Permission relayed by someone else does not pass.

Not for: creators who want an app instead of code. We could not read Speechify's consumer Studio pricing page, so its Studio plans are not covered here.

Murf (Stork listing) is a voiceover studio with a large stock voice library. Its voice cloning page is a Contact Sales form; it asks whether you have under or over 30 minutes of recorded audio, and its FAQ says professional cloning generally needs multiple samples. A price is not stated, and Murf's pricing page did not load as text for us.

Not for: an individual creator who wants to clone a voice today. Pick Murf if your team already uses its stock voices and wants a brand voice handled through a sales process.

If you have a GPU and want the model on your own machine, Fish Speech is the most active open project we found. Its README says 10 to 30 seconds of reference audio is enough to reproduce a voice with its S2-Pro model, and a newer Drama 3 model also supports cloning. The code and weights are under the Fish Audio Community License, with S2-Pro weights under a Fish Audio Research License; whether commercial use is allowed is not stated in the README, so read the LICENSE file before you ship anything.

Chatterbox, Resemble AI's MIT-licensed model, was a common pick in 2025. Its README now says that as of August 2026 it "is retired research and is no longer maintained". The code still works as an archive and stamps every output with an inaudible watermark.

Not for: anyone without a GPU or the patience to set up Python. Open models also come with no consent check at all, so the legal duty sits entirely with you.

Cloning your own voice is fine everywhere on this list. Cloning someone else's is where accounts get closed. Here is what each vendor's policy says, in its own words:

• ElevenLabs. Its prohibited use policy bans cloning or impersonating the voice of "any living individual" without "explicit and informed consent or other valid authorization", bans recordings of anyone under 18, bans impersonating candidates or officials, and forbids removing its watermarks. Instant clones ask you to confirm you have the right and consent to clone the voice.
• Fish Audio. Its Terms do not mention cloning by name. They bar content that violates anyone's rights, require you to hold all rights to what you upload, and encourage you to disclose AI-generated content. Uploads may be used to train its models.
• Cartesia. Its terms bar posting "any other person's voice without that person's express permission", including the voices of deceased people and political candidates, and bar output that violates privacy or publicity rights. Inputs may be used for training unless you opt out through its form.
• Descript and Speechify. Both require a recorded consent statement from the person being cloned. Descript will not clone deceased people, non-consenting people, or AI-generated audio.

In practice: get written permission from anyone whose voice you clone, keep the recording of their consent, say when audio is AI-generated, and never clone a public figure for anything that could pass as real. The vendor's rules are a floor; local publicity and privacy laws can be stricter.

• You want to narrate videos or courses in your own voice: ElevenLabs Starter for an instant clone, Creator for a trained one.
• You want to try cloning for free: Fish Audio, and upgrade before any commercial use.
• You are building a voice agent or app: Cartesia Pro at $5, or the Speechify API if a verified consent recording matters to you.
• You edit podcasts and want to fix words by typing: Descript Hobbyist.
• You want a brand voice built for a company: Murf through its sales team, or an ElevenLabs Pro or Business plan.
• You want full control and no subscription: Fish Speech on your own GPU, after reading its license.

Already on ElevenLabs and comparing it against the field for text-to-speech as a whole? Our ElevenLabs alternatives comparison covers the wider list.

What is the best AI voice cloning tool in 2026?

For cloning your own voice for narration and content, ElevenLabs is the most complete: instant cloning from Starter at $6 a month and a trained Professional Voice Clone from Creator at $22. Fish Audio is the cheapest way to start, Cartesia suits developers, and Descript suits podcast and video editors.

Can I clone my voice for free?

Yes, on Fish Audio's free plan, which lists voice cloning with 8,000 credits a month. Its terms limit free output to personal, non-commercial use. Descript's free plan also includes limited custom voice clones. ElevenLabs, Cartesia and the Speechify API do not include cloning on their free plans.

How much audio do I need to clone a voice?

For an instant clone, little: Cartesia says 10 seconds is enough to start, and ElevenLabs recommends 1 to 2 minutes of clean audio. A trained or professional clone needs far more: at least an hour on ElevenLabs, and 30 minutes or more on Cartesia.

Is it legal to clone someone else's voice?

Only with their explicit permission, and vendor rules are strict. ElevenLabs bans cloning any living person without explicit and informed consent and lets you make a Professional clone of your own voice only. Speechify and Descript require a recorded consent statement from the person being cloned. Local publicity and privacy laws can add further limits.

Is Play.ht still available?

No. Play.ht shut down and its domain no longer loads. ElevenLabs, Fish Audio and Cartesia are the closest replacements for its voice cloning.

Is there an open-source AI voice cloning model?

Fish Speech is the most active open project we found; its README says 10 to 30 seconds of reference audio is enough with the S2-Pro model. It uses Fish Audio's own community and research licenses, so check commercial terms first. Resemble AI's Chatterbox is MIT-licensed but was retired in August 2026.