The short answer
To clone your voice with AI, use ElevenLabs on the Starter plan or above: open Voices, click the plus icon, choose Instant Voice Clone, upload or record about 1 to 2 minutes of clean audio, confirm you have the right and consent to clone the voice, and click Save voice. The free plan has no voice cloning and no commercial licence. Professional Voice Cloning, from the Creator plan, needs at least 30 minutes of audio and only works for your own voice.
What you need first
- The right plan. ElevenLabs is the tool this guide uses, because its cloning steps and rules are documented in detail. Its free plan gives 10,000 credits a month but no Instant or Professional Voice Cloning and no commercial licence. Instant Voice Cloning starts on Starter, which also adds the commercial licence; Professional Voice Cloning starts on Creator with one slot. See the official pricing page for current prices.
- Consent. If the voice is not yours, you need the owner's explicit, informed permission before you start. More on this below.
- A quiet room and a decent microphone. A headset or USB microphone in a room with soft furnishings beats an expensive microphone in an echoey kitchen.
- A script to read. Two minutes of text in the style you want the clone to speak: calm narration if you will use it for training videos, lively if it is for social clips.
How to record samples that clone well
ElevenLabs' own guidance is blunt: the clone copies everything it hears, including pace, accent, breathing and background noise. So the recording matters more than any setting.
- Length. Aim for about 1 to 2 minutes. Record at least 1 minute and stay under 3 minutes, because extra length adds little and can make the clone worse. Each file must be at least 5 seconds, and all files together at least 10 seconds.
- One speaker, no music, no echo. Only your voice, loud and clear.
- Consistency. Same microphone, same distance, same tone across every sample. Big swings in pitch or volume give less predictable results.
- Level. ElevenLabs suggests roughly -23 dB to -18 dB RMS with a true peak of -3 dB. In plain terms: not too quiet, never clipping.
- Format. MP3 at 128 kbps or higher works. WAV usually does not improve results and can cause upload problems.
- Perform the way you want it to sound. The clone reproduces the delivery it hears, so read in the style you will use.
How to create an Instant Voice Clone in ElevenLabs, step by step
- Sign in to ElevenLabs on a Starter plan or above.
- Select Voices in the left sidebar, then click the plus icon.
- In the window that opens, select Instant Voice Clone.
- Upload your audio files, or record directly, following the on-screen instructions.
- Give the voice a clear name and labels, for example "Sam, calm narration, UK English".
- Confirm that you have the right and consent to clone the voice, then click Save voice.
- To use it, go to Voices, click My Voices, then Use voice, and type or paste your script into text to speech.
With Eleven v4, the flagship model since 28 September 2026, you can steer delivery with bracketed audio tags such as [excited], [whispers] or [short pause]. The Enhance button in the app suggests tags for your text. SSML break tags are not supported in v4 and v3, so use tags or punctuation for pauses instead.
When to move up to a Professional Voice Clone
An Instant clone is good enough for most voiceovers. If it misses your accent or you need audiobook-length reads, ElevenLabs points to Professional Voice Cloning, which fine-tunes a model on far more audio: 30 minutes is the bare minimum and closer to two to three hours gives the most accurate result. It needs the Creator plan or above, and it only works for your own voice. Every Professional clone goes through a verification step to confirm the voice is yours, so you cannot make one of a colleague even with their permission. They create and verify it on their own account and share it with you.
The consent rules, in plain English
From ElevenLabs' Prohibited Use Policy, updated on 9 October 2026, and its Safety page:
- You may not clone or impersonate the voice of any living person without their explicit and informed consent or other valid authorisation.
- You may not submit recordings of anyone under 18, or impersonate political candidates or elected officials without authorisation.
- ElevenLabs says it blocks cloning of celebrity and other high-risk voices, and offers a free AI Speech Classifier to check whether a clip came from ElevenLabs.
- Violations can lead to removed content and suspended or terminated accounts.
The working rule: only clone your own voice, or a voice whose owner has given clear written permission for that specific use, and keep the permission on file. If a client asks for a famous voice, the answer is no. Tell your audience when a voice is synthetic, too; in many workplaces that is already policy.
A worked example: weekly training updates in your own voice
Say you record a two-minute internal update every Friday and want to stop re-recording when you fluff a line. Record two minutes of yourself reading last month's update in your normal delivery, in the same room with the same headset. Create an Instant clone on Starter as above and name it clearly. Each Friday, paste the script into text to speech, choose your clone and Eleven v4, and add one tag at the start of a line where you want warmth or emphasis. Listen once, regenerate only the paragraph that sounds wrong, and download. Because Starter includes a commercial licence, you can also use the same voice on a public product video. Before you do, switch off the training setting described below.
Common mistakes and how to fix them
- The clone sounds robotic or has a hiss. The samples had noise or echo. Fix: re-record in a quieter room, or run a noise remover first, then create a new clone. You cannot change a clone's tone or accent after creation.
- More audio made it worse. Fix: stay between 1 and 3 minutes for an Instant clone; use Professional cloning if you have long recordings.
- Each clone sounds slightly different. That is normal: cloning the same audio twice gives slightly different voices. Fix: keep your original samples so you can recreate the one you liked.
- Expecting to export the voice. You cannot download a clone as a file; it stays in your account.
- Cloning in an unsupported language. Cloning works in the languages supported by Flash v2.5 and Multilingual v2. Fix: check the language list first.
- Leaving training on. ElevenLabs may use your inputs to improve its models unless you opt out. Fix: click your profile icon (top right), then Terms and privacy, then Data use, switch off "Improve the models for everyone" and click Update your choice.
Which tool should you use?
- ElevenLabs for the best control over a standalone voice, commercial voiceovers and dubbing. Start on Starter.
- HeyGen if the voice is for an avatar video of you. Its free plan includes 1 voice clone, and digital twins require an on-camera consent video from the person shown.
- Runway includes 1 custom voice clone on its Pro plan, useful if you already make video there.
For more options, read the best AI for voiceovers, ElevenLabs alternatives and ElevenLabs vs Descript.
Prompts to copy
Two-minute sample script
Write a 2-minute script (about 280 words) for me to read aloud when recording samples for a voice clone. I will use the clone for [use, such as training videos]. Match this style: [calm, warm, energetic]. Include a mix of short and long sentences, a few numbers and a question, and no tongue-twisters.
You get: A natural reading script that covers the range your clone needs
Rewrite for the ear
Rewrite this text so it sounds natural when read aloud by an AI voice: [paste text]. Use short sentences, contractions and punctuation that marks pauses. Spell out numbers, dates and acronyms as they should be spoken.
You get: A script that the clone reads without stumbling
Add audio tags sparingly
Add Eleven v4 audio tags to this script, using at most one tag per paragraph from this list: [warm], [serious], [excited], [whispers], [short pause]. Put each tag at the start of the line it affects. Script: [paste script]
You get: A tagged script with controlled, not over-acted, delivery
Consent form for a colleague
Draft a short plain-English consent form for [name] to allow [company] to create and use an AI clone of their voice for [specific use] until [date]. Include who may use it, where it may appear, how they can withdraw consent and what happens to the clone and samples if they do.
You get: A consent record to adapt with your legal or HR team
Two-speaker dialogue
Turn this content into a two-person dialogue for text to speech, under [number] words, formatted with each line starting 'Speaker 1:' or 'Speaker 2:'. Speaker 1 is [role], Speaker 2 is [role]. Content: [paste]
You get: A dialogue script in the format ElevenLabs expects for multiple speakers