AI Voice Cloning: How to Clone Your Voice and Use It Everywhere

A clone learns how a voice sounds, not what it knows. From one sample it picks up the pitch, the texture, the accent and the general pace of the speaker, then reads new text with them. It does not learn your jokes, your pronunciation of a rare name you never said in the sample, or the way you sound when you are furious. It reads; it does not act.

When a line needs a real performance, record it yourself and, if you want a different voice on it, run it through Voice Changer, covered below. When a line only needs to be read, the clone saves you the booth.

The clone can only be as good as the sample. A phone is enough: the sample on our voice cloning page is one take on a phone of about fifteen seconds, and every other read on that page was cloned from it. What matters is the room and the voice, not the microphone.

• At least 10 seconds of speech. The studio refuses a shorter file before anything is charged. Half a minute of steady talking gives the clone more of your voice to work from.
• One speaker. No second voice, no interviewer, no laughter from across the room.
• No music and no noise. Turn off the fan and the television, and record away from traffic. A small room with soft furnishings, a closet full of clothes for instance, beats a large tiled one.
• Your normal voice. Talk the way you will want to sound: if the clone will read ads, read something lively; if it will narrate, read something calm.
• A steady distance. Hold the phone about a hand's width from your mouth and keep it there; moving it changes the sound mid sentence.
• No long silences. Trim the start and the end so the file is speech from the first second.

Most failed clones come from the sample, not the model. If a read sounds like someone else, record a cleaner sample before you change anything else.

Zero-shot means there is no training run. You do not upload hours of audio or wait for a custom model: the voice is made from your one sample in seconds, the first time you run a read. Nothing is made from the sample when you save it, which is why saving costs nothing.

1. Open Voice Clone in the voice studio and add a voice: drop in a recording, or record one in the browser.
2. Name it. It saves to your account and appears in the studio's voice list.
3. Type the line you want read, up to 1,000 characters a read, and pick the language the text is in.
4. Run the read. The audio comes back in that voice, and credits settle only when the file arrives; a read that fails is not charged.
5. Listen once on headphones and once on a phone speaker. If it sounds right on both, it is ready to use.

The voice clone tool page does the same from one screen, if you want to try a sample before you open the studio. If the main engine cannot take a read, a backup engine reads it from the same sample and the result says so; a voice is never swapped without a word.

Three voices: your clone, a changer for your own recordings, and a preset voice for when the line should not be you. The table comes from our model catalog, so it says what each one does in the studio today.

• ElevenLabsVoice CloneReads in a voice you upload
• ElevenLabsVoice ChangerRe-speaks a recording in another voice
• ElevenLabsElevenLabs v3Eleven v3, direction tags, 21 presets

ElevenLabs v3 reads in preset voices and takes direction tags such as [whispers] or [laughs] in the text, which suits characters and voices that are not yours. Our ElevenLabs page and text to speech page play its presets.

A script written for the page sounds stiff when it is read aloud, even in your own voice. Write the way you talk, and the clone will sound more like you.

• Short sentences. One idea each. A clone handles a comma as a small pause and a full stop as a longer one, so the punctuation is your timing.
• Words you would say. Contractions, plain verbs, no list of three adjectives where one will do.
• Numbers the way you want them said. Write twenty twenty-six or two thousand and twenty-six if the difference matters to you; otherwise digits read fine.
• Names spelled the way they sound. A rare name or a brand the sample never said can come out wrong; respell it phonetically and read it back.
• One paragraph per read. A read takes up to 1,000 characters, a long paragraph or two. Split a longer script at the paragraph breaks and join the reads in your editor, with the same voice and the same language for each.
• Read it aloud yourself first. If you trip on a sentence, the clone will make it sound odd too.

Keep a tone consistent with the sample. A clone made from a relaxed sample reads a hard sell less convincingly than one made from an energetic sample. If you need both, save two voices from two samples and pick per script.

A read is an audio file, so it goes anywhere audio goes. These are the routes on DreamClip that take your voice further, each with the input it needs.

Ads

For an ad, a clone lets you rewrite the read after the shoot: a new offer, a new hook, a shorter cut. Keep each line short, render a few versions, and pick by ear. The read itself is the same cost whether the line is good or bad, so the writing is where the savings are. Before an ad runs, read our terms: commercial use covers generations made on paid plans.

Dubbing

Video dubbing re-speaks the speech in a clip in another language, in a voice close to the original, and leaves the picture as it was shot, so the lips still move as they did. When the mouth has to match the new words, use the second route instead: read the translated line with your clone, then run Lip Sync on the clip with that read.

Our guide to animating old photos uses the same avatar route for a family portrait, with the care that subject needs.

A clone reads text. A voice changer keeps a performance. Voice Changer takes a recording you made, audio or video, and re-speaks it in another voice with your timing, your pauses and your emphasis intact. Pick one of its preset voices, or one of your saved voices.

• Act a character yourself, then give it a voice that is not yours.
• Record a line on a bad day, then change it into your saved voice from a good one.
• Keep a colleague's delivery for a demo and put it in your own saved voice, with their agreement.

It takes recordings of up to five minutes and is priced by the second of the recording. A video comes back as a video with the new voice as its sound, so a talking clip stays in sync because the timing never changed.

Clone only a voice you have the right to use: your own, or one whose owner has agreed. Our terms forbid using the service to create content that impersonates real people without their consent, and we may remove content or suspend accounts that do. You are responsible for your uploads and for what you make with them.

• Your own voice: yours to clone and to use.
• A colleague, a client or a presenter: get their agreement in writing, and say what the clone will be used for.
• A public figure, a celebrity or a voice from a film: no.
• A relative who has died: only with the family's agreement, and keep the result within the family.

Your sample is stored in your own account. It is sent to the voice model only when you run a read, and only to make that read, and deleting the voice deletes the recording. You keep the rights to what you make, subject to the terms of the underlying model providers.

If someone asks you to clone a voice and cannot show that its owner agreed, the answer is no. A cloned voice saying words its owner never said can do real harm.

• Range. The clone reads in the mood of the sample; big emotion needs a recorded performance and Voice Changer.
• Rare words. Names and terms the sample never contained can come out wrong; respell them.
• Long scripts. A read takes up to 1,000 characters, so long pieces are several reads joined in an editor.
• A noisy sample. Room echo and background sound carry into every read; re-record rather than fight it.

ElevenLabs sells voice cloning directly on its own site, with its own plans, and Descript, the video and podcast editor, has AI voice features for correcting words in a recording, both as of October 2026. DreamClip is neither of them. Here the clone sits in the same studio as the video tools, so a read goes straight onto a talking avatar, a lip synced clip or a dubbed video without a second account.