Best AI voice cloning: which tool sounds most like you?

Contents

What is the best AI voice cloning tool?

The best AI voice cloning tool is ElevenLabs. It does both kinds of cloning well, an instant clone from a minute or two of audio and a professional clone trained on 30 minutes or more, it starts at $6 a month, and a clone speaks every language its models support. If you want one AI voice cloner and nothing else, that is it.

But “best at cloning” splits into different jobs once you start, so here is the decision in five lines.

• Cloning your own voice to fix a recording? Descript, which patches a line in your voice by typing, inside a real editor.
• Building a live voice agent? Cartesia, with first audio in under 90 ms advertised and instant cloning from $5 a month.
• The clone has to talk on camera? HeyGen, which turns a 30-second sample into a lip-synced, translated video.
• Film, TV or games, where an actor’s performance has to carry? Respeecher, which converts one person’s speech into another’s voice.
• Refuse to pay a cent? An open-source model like Chatterbox, if you can run it yourself.

One honest note before the list. I score ElevenLabs 4.7 in our full ElevenLabs review, and I would still pick it first for cloning quality. The tools below are not here because they clone better. They are here because they clone differently, cheaper, faster or for free, and one of those differences is probably why you are reading this.

How did we pick these voice cloners?

I ranked eight voice cloning tools on fit for a specific job, weighing similarity, audio needed, price to clone, languages, response time and consent, with every price re-checked on November 4, 2026.

Voice cloning is a narrow promise with a wide set of jobs hiding inside it. “Copy a voice and make it read a script” covers a podcaster patching one flubbed line, a developer wiring a voice into a phone agent, a studio dubbing a film, and a hobbyist who wants their own voice for free. Those are different buyers, and they do not want the same tool.

So I ranked on fit, not on a single quality number. Every price, plan limit and sample length below was re-read on the vendor’s own pricing page or documentation.

I have been clear about where the read comes from. I cloned my own voice in ElevenLabs and in Cartesia in June 2026, and both clips are on this page. Descript, HeyGen and Murf were tested hands-on for their own reviews, but I have not trained a clone in any of the three. I have not tested Fish Audio, Respeecher or the open-source models at all, and their sections say so; those reads come from the vendors’ docs, pricing pages and license files.

Criterion	What I weighed
Similarity	How close the clone sounds to the source, and whether a professional mode exists
Audio needed	Seconds for an instant clone, or 20+ minutes for a professional one
Price to clone	The cheapest plan where you can actually clone a voice
Languages	How many languages a single clone can speak
Response time	Time to first audio, for anyone building a live agent
Consent and rights	Verification before cloning, and whether you can sell the output

Four names you might expect are missing, and the reasons are worth stating. Resemble AI used to rank here for its watermarking, but its own voice-cloning page now says it has stopped offering voice generation as a commercial product; it sells deepfake detection instead, and its open model, Chatterbox, appears in the open-source section. Speechify no longer lists cloning on its Studio pricing, and its old cloning page redirects to the homepage; cloning survives only in its developer API.

WellSaid Labs builds its library from licensed voice-actor recordings and does not advertise self-serve cloning. And Play.ht, still on many older lists, no longer resolves, so recommending it would send you to a dead page.

Instant vs professional voice cloning: which do you need?

You need instant cloning to test, prototype or dub a short clip, and professional cloning when the clone carries real weight. Most cloud cloners offer both modes, and picking the wrong one is a common reason people say “the clone didn’t sound like me.”

Instant cloning copies a voice from a short sample, typically 10 seconds to 2 minutes, in a few seconds of processing. It is good in short bursts and slips on the hard parts: unusual words, strong emotion, long passages, and accents the model has not heard much. ElevenLabs’ own documentation warns that instant clones “may struggle with uncommon accents or unique voices.”

Professional cloning trains a dedicated model on far more audio, from 20 minutes at HeyGen to 30 minutes or more at ElevenLabs, which calls 3 hours optimal. Training takes longer and usually costs a higher plan or an add-on. The payoff is stability over a long read, which is what an audiobook or a year-long brand voice needs.

Whatever a product calls itself, voice cloner, AI voice replicator or voice clone generator, it runs one of these two modes.

The trap is the source audio, not the tool. A quiet, consistent 30-minute recording clones better than two noisy hours with a fan humming in the background. Every tool below inherits that rule.

The measure	Instant cloning	Professional cloning
Sample needed	10 seconds to 2 minutes	20 minutes to 3 hours, clean
Time to clone	Seconds	Longer training run
Similarity	Good, slips on accents and long reads	Built for long, stable reads
Best for	Prototypes, short dubs, agent personas	Audiobooks, brand voice
Cheapest entry	Cartesia $5, ElevenLabs $6, HeyGen free	Fish Audio $132/yr, ElevenLabs $22, Cartesia $49, HeyGen $29 plan + $29 add-on

How do the best voice cloning tools compare?

ElevenLabs leads on realism, Cartesia on response time, and Fish Audio and HeyGen on entry price. Here is how all eight compare as voice cloning software; “Cloning from” is the cheapest plan where you can clone a voice, checked November 4, 2026.

Tool	Best for	Instant clone sample	Professional clone	Cloning from	Languages	We tested cloning?
ElevenLabs	Overall realism	1–2 min recommended	30+ min, Creator $22	$6/mo Starter	32 (Flash v2.5), 70+ (v3)	Yes, June 2026
Descript	Cloning your own voice	Not published	Not offered separately	$24/mo Hobbyist	Not stated	No (editor tested)
Cartesia	Real-time agents	~10 seconds	Startup $49 (2 slots)	$5/mo Pro	44	Yes, June 2026
Fish Audio	Cheap cloud cloning	10–15 seconds	1 slot on Plus	Free (personal) / $132/yr	8 listed, “30+” claimed	No
HeyGen	Clones in video	30 seconds	20+ min, $29/mo add-on	Free (1 clone)	175+ claimed	No (video tested)
Respeecher	Film and games, speech-to-speech	Not published	Custom projects	Custom quote	Not stated	No
Murf AI	Teams already on Murf	Custom	Custom	Enterprise add-on	30+ (studio)	No
Open source	Free and self-hosted	From 6 seconds (XTTS)	Fine-tune yourself	Free (you host)	6 to 23 by model	No

1. ElevenLabs — best overall voice cloning

E

ElevenLabs is the default most buyers compare against, and it is the AI voice cloner I would hand a friend who asked for a single answer. The reason is not one feature. It is that both cloning modes are good, cheap to reach, and wired to a voice engine that already sounds natural before you clone anything.

Instant Voice Cloning builds a usable clone in seconds. The marketing page leads with a 10-second recording, but the documentation is more useful: it recommends 1 to 2 minutes of clean audio, asks for at least a minute, and warns against going past 3. Professional Voice Cloning is the real draw. Feed it 30 minutes or more, with 3 hours called optimal, and it trains a dedicated model after a voice verification step that checks you are allowed to clone that voice.

Here is my own voice, cloned with ElevenLabs in June 2026, reading a script I never recorded. Judge the fidelity yourself.

That is the engine your clone runs on. To hear its range on voices I did not clone, here is stock text-to-speech and the same model crossing into Spanish:

Pricing is where you need to read the tiers carefully, because the two modes sit on different plans. The free plan gives 10,000 credits a month, roughly 10 minutes of speech, with no cloning and no commercial license. Instant cloning and the commercial license arrive on the $6 Starter plan. Professional cloning starts on the $22 Creator plan, which includes one professional voice; Scale at $299 includes three.

The catch is the one every ElevenLabs user lands on: credits bill every regeneration, so every retake of a line is paid for again. The voices are not the complaint; the bill is. Our ElevenLabs pricing guide works through what the credits cost once you are regenerating lines in volume.

The measure	ElevenLabs
Instant clone	1–2 min recommended, from $6/mo Starter
Professional clone	30+ min (3 h optimal), from $22/mo Creator
Languages	32 on Flash v2.5, 70+ on Eleven v3
Speed	~100 ms median on Flash v2.5, before network time
Consent	Voice verification before a professional clone

Read the full ElevenLabs review for where the credits really go, or our ElevenLabs alternatives guide if the pricing is the dealbreaker.

2. Descript — best for cloning your own voice

D

Descript clones for a completely different reason than everyone else, and that reason is the whole point. Its voice cloning, long known as Overdub and now sold as AI Speech with custom voice clones, exists so you can fix a recording by typing, not to mass-produce narration from a stranger’s sample. If your real job is editing podcasts or talking-head video, this is the cloner that lives where you already work.

The workflow is the trick. Descript turns your recording into a transcript, you edit the media by editing the text, and when you need to patch a flubbed line, it speaks the correction in your cloned voice instead of sending you back to the mic. Its Regenerate feature goes a step further and rebuilds a stretch of your own audio to match the tone around an edit.

I tested Descript’s editor for our review; I have not published a clone sample from it, so the quality call here leans on its documentation and on what it is designed for. Descript requires a consent recording before it clones your voice, which is the right call on a feature this easy to misuse. And the honest framing is that it is a patch tool, not a narration engine. For reading an hour of new script in a cloned voice, ElevenLabs is the better tool.

Custom voice clones are included from the Hobbyist plan at $24 a month, or $16 a month billed annually, with Creator at $35 and Business at $65; the free plan lists cloning as “limited.” The thing to watch is Descript’s two meters, media minutes and AI credits, which heavy users burn through faster than they expect.

The measure	Descript	ElevenLabs
What it clones	Your own voice	Any voice you can verify
Cloning style	Patch a line by typing	Full narration
Cloning from	$24/mo Hobbyist ($16 annual)	$6/mo Starter
Best for	Editing recordings	Generating from scratch

Read the full Descript review for how the two meters actually meter.

3. Cartesia — best for cloning a voice into a live agent

C

Cartesia is the developer’s cloner, and it competes on one number above all: response time. Its Sonic models are built for live conversation, and Cartesia advertises first audio in under 90 milliseconds. In a phone agent, that is the difference between a reply and an awkward pause.

I cloned my own voice in Cartesia in June 2026, from roughly 10 seconds of audio. To my ear it sounds like me rather than a rough approximation. Here is the clone reading a line I never recorded, so you can set it against the ElevenLabs clip above; both are the same voice.

The pricing is low. A free tier gives 20,000 credits a month, the $5 Pro plan lists instant voice cloning and a commercial license, and professional cloning arrives on the $49 Startup tier with two clone slots. One date matters here: in my June test, instant cloning worked on the free plan, but the November 2026 pricing page lists it from Pro. Plan for $5.

What you give up is polish and hand-holding. Cartesia is squarely a build-it-yourself tool, with a playground that hands you the API call rather than a point-and-click studio. For narration you assemble by hand, ElevenLabs is the nicer place to work. Sonic covers 44 languages, which is plenty for most agents.

The measure	Cartesia	ElevenLabs
Instant clone from	$5/mo Pro	$6/mo Starter
Professional clone	$49/mo Startup	$22/mo Creator
Advertised speed	Under 90 ms to first audio	~100 ms median (Flash v2.5)
Best for	Real-time agents	Polished narration

Read the full Cartesia review for the playground and the per-minute agent pricing.

4. Fish Audio — best low-cost cloud cloner

F

Fish Audio is the cheapest route to a professional voice slot on this list, and it comes with an unusual bonus: the company also publishes open model weights. You can use the hosted app like any other cloner, or take the model in-house later. I have not tested Fish Audio, so this section is built from its plan page, homepage and license file.

Cloning is quick on paper. The FAQ says as little as 10 seconds of audio makes a clone, while the cloning section talks about a 15-second clip; budget for the longer one. The platform also hosts a community library that Fish Audio puts at over 2,000,000 voices.

The plan page is where it gets interesting. The free plan gives 8,000 credits a month, and at the roughly 600 to 625 credits per minute the FAQ quotes, that is about 13 minutes of speech, for personal use only. Plus costs $132 a year with 250,000 credits, 10 private voices and one professional voice slot, plus commercial use. That works out to about $11 a month for a professional voice slot on annual billing, against $18.33 a month for ElevenLabs Creator billed annually. Pro is $900 a year with 2,000,000 credits and five professional slots. The API is billed separately, pay as you go.

Two things to weigh before you commit. The language claims do not agree with each other: the FAQ lists eight languages, while a feature block promises “30+”. And I found no consent or verification wording on the homepage, which puts the burden on you rather than the product.

The measure	Fish Audio	ElevenLabs
Sample needed	10–15 seconds	1–2 min (instant)
Entry price for commercial use	$132/yr Plus	$6/mo Starter
Professional voice	1 slot on Plus	From $22/mo Creator
Open weights	Yes, under a community license	No

Pick Fish Audio if you want cheap cloud cloning now and the option to self-host later.

5. HeyGen — best for cloning into video

H

HeyGen clones a voice so it can drive a video, and that framing is the whole pitch. Where the others hand you an audio file, HeyGen bakes the clone into a text-to-video editor: paste a script, and it generates the narration in your voice, syncs an avatar’s lips to it, and can translate the result. If the end product is a talking-head video, this removes the most steps.

Cloning is light on input. HeyGen’s cloning page says a clear 30-second sample is enough for an instant clone, and a professional clone trains on 20 minutes or more. HeyGen verifies voice ownership before training and asks for written consent for any third-party voice. The language count it quotes is 175-plus.

I tested HeyGen’s avatar video on the free plan in July 2026, but not its voice cloning. One thing from that test is worth knowing: the in-app upgrade screen then listed voice cloning among the features the free plan withheld. Today’s pricing page lists “1 voice clone” on Free, so check the app before you plan around it.

The paid plans start at $29 a month for Creator ($24 billed annually), then $49 for Pro and $149 plus $20 a seat for Business. Professional cloning is an add-on on top of a paid plan: $29 a month or $288 a year, per voice. It is the wrong tool if all you want is an audio file, and the right one if the clone was always meant to talk on camera.

The measure	HeyGen	ElevenLabs
Output	Cloned voice in a video	Cloned voice (audio)
Instant clone	30 seconds, free plan lists 1	1–2 min, from $6
Professional clone	20+ min, $29/mo add-on per voice	30+ min, from $22/mo
Best for	Talking-head and translated video	Audio-first projects

Read our HeyGen review for what the free plan really allows, and HeyGen pricing for the credit math.

6. Respeecher — best for film, games and speech-to-speech

R

Respeecher clones from a performance, not from a script. Its core product is speech-to-speech: an actor performs a line with the timing and emotion the scene needs, and the system converts that performance into the target voice. That keeps every breath and pause a text-to-speech clone has to guess at, which is why studios use it.

The credits Respeecher lists are the argument for it. They include the younger Luke Skywalker voice in The Mandalorian, Hungarian pronunciation for the leads in The Brutalist, and a synthetic Richard Nixon for In Event of Moon Disaster. I have not tested Respeecher; everything here comes from its own site.

It is not a self-serve cloner. Custom voice work is quoted through the sales team, and Respeecher lists consent mechanisms and verification processes among its commitments without detailing them publicly. The self-serve parts are a Voice Marketplace of 40-plus licensed voices with a Pro Tools plugin, and a real-time text-to-speech API priced at $2 an hour, pay as you go.

The measure	Respeecher	ElevenLabs
Core method	Speech-to-speech from an actor	Text-to-speech from a script
Cloning price	Custom quote	From $6/mo
Self-serve option	Marketplace voices, $2/hour TTS API	Full self-serve
Best for	Film, TV, games, dubbing	Creators and apps

Pick Respeecher if a cloned voice has to carry an actor’s performance, not just read text.

7. Murf AI — best if your team already lives in the studio

M

Murf earns a spot the way a strong all-rounder makes a specialist list: not because it clones best, but because if your team already pays for it, a custom voice can bolt onto the studio you use. It is a text-to-speech tool wrapped in a production studio, with a video timeline, music, subtitles and team folders, plus 200-plus stock voices across 30-plus languages and accents.

The cloning is where Murf asks you to pay up. Custom voice clones appear only on the Enterprise plan, as an add-on, with no public price. The self-serve plans, Creator at $19 a month and Business at $66 a month on annual billing, do not include them. Set against Cartesia’s $5 or ElevenLabs’ $6, a sales-gated clone with no public price is the hardest one here to budget for. The Murf pricing guide covers what the published plans do include.

I tested Murf’s Studio voices for our review in June 2026; I have not trained a Murf clone. So the verdict is conditional. If you need to clone a single voice and nothing else, the tools above with public prices are the simpler buy. If your team already runs training content through the studio and wants one vendor for voice, video and collaboration, a custom clone is a reasonable add-on rather than a reason to switch.

The measure	Murf AI	ElevenLabs
Cloning available at	Enterprise add-on (Contact Sales)	$6/mo Starter
Public cloning price	None	Yes
Surrounding studio	Video, music, folders	Voice-first
Best for	Teams already on Murf	Cloning on any budget

Read the full Murf AI review for where the studio earns its price and where it does not.

8. Open source — best for free, private cloning on your own GPU

If the real objection is paying anything at all, skip subscriptions. Open-source voice models are a serious option for running your own clone; their maintainers pitch quality close to the paid tools, which I have not verified. The catch is that “open” does not always mean “free to sell,” and the licenses differ more than most lists admit. I have not tested these models; the details below come from each project’s repository and license.

• Chatterbox is the pick if you need commercial use. It is MIT-licensed, its multilingual model covers 23 languages, and every file it generates carries an imperceptible Perth watermark, which is a rare piece of built-in responsibility.
• OpenVoice is the lightweight option, MIT-licensed for commercial and research use, with six languages built in and fine control over tone and accent.
• Coqui XTTS v2 clones from a 6-second clip in 17 languages, but its Coqui Public Model License is non-commercial. Fine for a personal project, wrong for client work.
• Fish Audio’s open weights allow commercial use only while your annual revenue stays under $500,000, and never for running a hosted voice service.

The honest catch is the one a subscription hides: you trade the bill for setup time and hardware. Most of these want a GPU, so you own one or rent one by the hour. There is no support line when something breaks and no polished editor.

There is also a quieter advantage that matters for cloning specifically: privacy. Running the model yourself means the source voice never leaves your machine, which for a personal or sensitive voice is worth more than any feature on a pricing page.

The measure	Chatterbox	OpenVoice V2	Coqui XTTS v2
License	MIT	MIT	Non-commercial (CPML)
Languages	23 (multilingual model)	6	17
Watermark	Yes, every file	No	No
Commercial use	Yes	Yes	No

Which voice cloning software sounds most like the real person?

A professional clone from ElevenLabs, trained on 30 minutes or more of clean audio, is my pick for the closest match you can buy self-serve. That call rests on ElevenLabs’ own documentation for professional clones and on my own ElevenLabs clone; I did not record which mode that June clone used. I have not run a blind listening test across every tool here, and no vendor page I read publishes one I would trust.

What I can give you is the comparison most guides skip: the same voice, cloned by two tools. The ElevenLabs clip and the Cartesia clip above are both clones of my voice, made in June 2026. Play them back to back and judge which one you would ship. My own read, from our ElevenLabs vs Cartesia comparison: both nail the identity, and ElevenLabs (clone mode not logged) carries more of the rise and fall that makes a long read sound human. On a one-line agent greeting the gap disappears.

Similarity depends on four things more than on the brand:

• The mode. Instant clones are convincing on a sentence; a professional clone holds up over an hour.
• The source audio. Clean, consistent, one microphone, no music bed. Noise in means noise out.
• The accent. ElevenLabs’ own docs say instant clones can struggle with uncommon accents, and recommend a professional clone if yours does not carry.
• The language. Speechify’s API docs say a clone speaks other languages with the accent of its recording, so record in the language you plan to generate.

Which voice cloning stack is best for real-time conversations?

Cartesia is the best voice cloning stack for real-time conversation, because response time is the whole product. ElevenLabs is the alternative when you also need its voice quality and agent platform. For a live agent, time to first audio is the number that decides whether a caller hears a reply or a stall.

Be careful with the numbers, because vendors measure them differently. The figures below are each vendor’s own, and most leave out network travel and your app’s processing.

Model	Advertised speed	What the number covers	Clone from
Cartesia Sonic	Under 90 ms to first audio	Vendor figure	10 s, $5/mo
ElevenLabs Flash v2.5	~100 ms median	Model only, excludes network	1–2 min, $6/mo
ElevenLabs Eleven v3 Conversational	~165 ms median	Model only, excludes network	Same voices
Chatterbox Turbo	No figure, “built for low-latency voice agents”	Self-hosted, depends on your GPU	Reference clip

Independent measurements run higher than the marketing. A public LinkedIn benchmark cited in our Cartesia review measured Sonic 3 at a 164 ms average that stayed under 300 ms on every call, which is still fast enough for natural turn-taking.

One related job is not covered by these numbers: speech-to-speech, where you record a performance and the tool swaps in a cloned voice. That is mainly a recorded workflow rather than a live call, and it is what Respeecher and the ElevenLabs voice changer are for.

Two practical rules. First, test from the region your callers are in, because the network can add more delay than the model. Second, clone with an instant clone for the agent persona, since a one-sentence reply rarely shows the gap a professional clone closes. For the full developer comparison, including pricing per character, see our best TTS API guide and ElevenLabs vs Cartesia.

How much audio do you need to clone a voice?

Instant voice cloning needs 10 seconds to 2 minutes of audio, and professional cloning needs 20 minutes to 3 hours, depending on the tool (checked November 2026). The numbers vary by tool, so here is what each vendor actually asks for.

Tool	Instant clone	Professional clone
ElevenLabs	1–2 min recommended, at least 1 min, no more than 3	30+ min, 3 hours optimal
Cartesia	About 10 seconds	Startup plan and up
Fish Audio	10–15 seconds	One slot on Plus
HeyGen	30 seconds	20+ minutes
Coqui XTTS v2	6 seconds	Fine-tune yourself
Speechify API	10–30 seconds, plus a consent phrase	Not offered

More audio only helps if it is good audio. Record in a quiet room, keep the same microphone and distance, speak in the style you want the clone to use, and trim long silences. If you are cloning for narration, read narration; if for an agent, talk like an agent.

Is there a truly free way to clone a voice?

Yes, but only self-hosted open-source models are free for commercial work. Chatterbox and OpenVoice are MIT-licensed, while HeyGen’s and Fish Audio’s free cloud tiers are for testing or personal use. “Free” means two different things here, and the gap between them catches people out.

The first kind is a free cloud tier. HeyGen’s free plan lists one instant clone, and Fish Audio’s free plan includes cloning with 8,000 credits a month, but Fish Audio limits free output to personal use. ElevenLabs’ free plan does not include cloning at all, and Cartesia now lists instant cloning from its $5 plan. These tiers let you judge a clone, not sell what it makes.

The second kind is genuinely free: an open-source model on your own hardware. Chatterbox and OpenVoice are MIT-licensed, so their output can be used commercially at no per-minute cost. The price moves from your wallet to your time and a GPU.

So for a test clone, use HeyGen’s or Fish Audio’s free tier. For ongoing, commercial, zero-cost cloning, Chatterbox or OpenVoice qualifies, as long as you will set it up. Free, polished and one-click at the same time does not exist yet.

Is AI voice cloning legal? Consent, verification and deepfakes

Cloning your own voice is legal, and cloning someone else’s is legal with their clear consent. Cloning a real person without permission is where it turns: it can break likeness and publicity rights, and a cloned voice used to deceive, say to fake an authorization or impersonate someone for money, crosses into fraud. “I found the audio online” is not consent.

The tools split on how seriously they enforce this. The stricter ones build it into the product rather than leaving it in the terms of service.

Safeguard	Who does it	What it does
Voice verification	ElevenLabs (professional clones)	Checks you control the voice before training
Ownership check and written consent	HeyGen	Verifies the voice; written consent for third parties
Spoken consent challenge	Speechify API	Speaker reads a one-time phrase that is matched to the sample
Consent recording	Descript	You record a consent statement before cloning your voice
Watermark on every file	Chatterbox (open source)	Marks generated audio as synthetic
Deepfake detection	Resemble AI	Flags cloned audio after the fact, from $0.035 a second

Resemble is the odd one out here. Having stopped selling voice generation, it now sells detection on a pay-as-you-go basis, so a legal or trust team can test a suspicious clip.

US regulators have also moved. In February 2024 the FCC ruled that AI-generated voices count as “artificial” under the Telephone Consumer Protection Act, so a robocall using a cloned voice needs the called party’s prior express consent, with limited exceptions.

The practical rule is short. Clone your own voice freely, get written consent for anyone else’s, and if your use is commercial or public, pick a tool that verifies. The friction now is cheaper than a likeness claim later.

How do you pick a voice cloning tool?

Pick by job: ElevenLabs for the closest clone, Descript for your own recordings, Cartesia for live agents, HeyGen for video. If that does not settle it, ask what you are cloning and where the voice will end up.

• You want the most convincing clone, full stop. ElevenLabs, with a professional clone on 30-plus minutes of clean audio.
• You want to fix your own recordings. Descript, which patches a line by typing in your cloned voice.
• You’re building a live voice agent. Cartesia, for sub-90 ms advertised response and cloning from $5. Our Cartesia pricing guide has the per-character math.
• You want cheap cloud cloning with an exit to self-hosting. Fish Audio.
• The clone needs to talk on camera. HeyGen, from a 30-second sample.
• An actor’s performance has to survive the clone. Respeecher, through speech-to-speech.
• Your team already runs on a voice studio. Murf, where a custom Enterprise clone bolts on.
• You refuse to pay and can run a GPU. Chatterbox for commercial use; XTTS only for personal projects.
• You want to test cloning free first. HeyGen’s free clone or Fish Audio’s free tier, then upgrade to whichever fits.

Notice what is not on the list: a tool that is cheapest, closest, fastest and free all at once. Pick the corner you care about most, and the choice falls out.

If you are choosing a voice for long-form narration rather than cloning one, see best AI voice for audiobooks; for stock voices across the whole category, best AI voice generator.

Final word

The best voice cloning software for most people is ElevenLabs, because it does instant and professional cloning well and starts at $6, but the winner changes the moment your job is narrower: Descript for your own recordings, Cartesia for live agents, HeyGen for video, Respeecher for film, and Chatterbox for free.

If you have not heard a clone of your own voice yet, start with an ElevenLabs instant clone. A minute of clean audio tells you more than any roundup can.