Aiodic

Back to the table

101 Creative tool El ElevenLabs Creative Tools

ElevenLabs

AI voice and audio.

What it is

ElevenLabs is a company, and a website of the same name, that turns written text into spoken audio that sounds like a real person. You paste in some words, pick a voice, press a button, and a few seconds later you have a recording you can download. It is one of the best-known tools for AI voice, and it is widely used by authors, podcasters, video makers, and businesses.

The company was started in 2022 by two friends who met at high school in Warsaw, Poland: Piotr DÄ…bkowski, who had been an engineer at Google, and Mati Staniszewski, who had worked at the data company Palantir. Their inspiration was a peculiarity of Polish television. Foreign films there are often not dubbed by a cast of actors. Instead, a single narrator reads every character's lines in the same flat voice over the top of the original sound. The two founders wondered whether AI could do far better: any film, in any language, with every character still sounding like themselves.

Here is what the tool does, in plain terms. The names of features change over time, but the abilities have stayed much the same:

How does it sound so human? Older computer voices, like early car navigation systems, worked by gluing together tiny recorded snippets of speech, which is why they sounded choppy. ElevenLabs uses deep learning, which means a system that has learned from a very large number of hours of real speech how written words turn into sound. That includes the things we never think about: where a speaker breathes, which word in a sentence gets the stress, how the voice lifts at a question mark and drops at the end of a thought. It doesn't replay recordings. It produces brand new audio each time, which is why pressing "generate" twice gives you two slightly different readings, the way an actor never delivers a line exactly the same way twice.

There is a free plan with a monthly allowance and paid plans with larger allowances and more features. Developers can also connect it to their own apps through an API, which is a way for one piece of software to use another.

Why it matters

It removes the biggest cost in audio. Hiring a professional narrator for a full-length novel commonly costs somewhere between one thousand and several thousand dollars, which put audio out of reach for most self-published authors. AI narration brings the price down to a small fraction of that. The same goes for podcast intros, course lessons, video voiceovers, and adverts. If you've avoided audio because of the expense, or because you dislike the sound of your own voice, that barrier is largely gone.

It lets a small creator work in many languages. Dubbing a video into Spanish, Hindi, or German used to require translators, voice actors, and an editor. Now one person can do a respectable version in an afternoon. For an entrepreneur, that can turn one market into several. It builds on AI translation and then adds the voice on top.

It gives voices back to people who have lost them. People with illnesses that take away speech can clone their voice from old recordings and keep speaking as themselves instead of through a generic computer voice. The same technology reads text aloud for people with sight loss or dyslexia. This is one of the clearest examples anywhere of AI doing something plainly good.

It raises a serious question about trust. Your voice is part of your identity. It is how your family recognises you on the phone and how some banks confirm who you are. If a minute of audio is enough to copy a voice, then a voice on the line is no longer proof of who is speaking. Anyone using these tools, and anyone who simply owns a phone, should understand both the promise and the risk.

Real-world examples

A member of Congress speaks in her own voice again

Jennifer Wexton, a member of the US House of Representatives from Virginia, was diagnosed with progressive supranuclear palsy, a rare brain disease that gradually took away her ability to speak clearly. In 2024, ElevenLabs built a clone of her voice from more than an hour of recordings of her earlier speeches. On 25 July of that year she used it to address the House of Representatives. It is believed to be the first time an AI-cloned voice was used to speak on the House floor. Until then she had been relying on a standard text-to-speech app with a robotic voice. "I cried happy tears when I first heard it," she said of the clone. For her, the technology didn't fake anything. It restored something the illness had taken.

Judy Garland reads The Wizard of Oz

In June 2024, ElevenLabs released an app that reads articles, documents, and e-books aloud. A few weeks later, the company announced agreements with the estates of several late Hollywood stars, including Judy Garland, James Dean, Burt Reynolds, and Sir Laurence Olivier, so that listeners could hear books read in those famous voices. Garland, who played Dorothy in the 1939 film, can "read" The Wonderful Wizard of Oz. What makes this example worth remembering isn't the novelty. It is how it was done: with permission from the people who hold the rights, and with payment. The voices are also restricted. You can listen with them inside the app, but you cannot use them to make content of your own. That is the model for using a famous voice legitimately.

The fake presidential robocall

In January 2024, in the last days before the presidential primary election in the US state of New Hampshire, thousands of voters received a phone call in what sounded like President Joe Biden's voice, telling them not to vote. It was a fake. Audio experts at the security company Pindrop and at the University of California, Berkeley examined the recording and concluded it had most likely been made with ElevenLabs, and the company banned the account responsible. Within weeks, the Federal Communications Commission, the US regulator for phone and broadcast services, ruled that automated calls using AI-generated voices are illegal unless the person called has agreed to them. It later fined the political consultant behind the calls 6 million dollars. ElevenLabs responded with a safeguard it calls "no-go voices", which blocks attempts to clone the voices of political candidates and other high-profile figures. Journalists have since shown that such blocks can sometimes be worked around, so no safeguard is perfect. It is a reminder that the feature that helps an author make an audiobook can also help a fraudster, and the difference is permission and intent.

An everyday case: an indie author's first audiobook

Picture Dana, who has self-published three mystery novels. Quotes for a human narrator came in at around three thousand dollars per book, more than the books have earned. She pastes her first chapter into ElevenLabs, auditions a dozen voices, and settles on a dry, slightly amused narrator that suits her detective. The first result is surprisingly good, but her detective's Welsh surname is mangled, and one tense line is read far too cheerfully. She fixes the name by spelling it the way it sounds, rewrites the punctuation on the tense line, and regenerates those two passages only. Over a couple of weekends she produces the whole book, listening to every chapter from start to finish. Before uploading it anywhere, she checks which audiobook shops accept AI narration and how they want it labelled.

Try it yourself

Menus and button names change from time to time, so treat these steps as a guide to the process and don't worry if the labels differ a little.

  1. Create a free account at elevenlabs.io. The free plan is enough for this exercise.
  2. Open the text to speech tool. It is normally the first thing you see after signing in.
  3. Paste in about 100 words of your own writing. Choose something with some feeling in it, such as the opening of a story, a product description, or the intro to a video. Flat, factual text won't show you what the tool can do.
  4. Audition several voices. Open the voice library and preview at least five. Don't take the first pleasant one. Ask yourself who would naturally say these words: what age, what accent, what energy?
  5. Generate the audio and listen with your eyes closed. Note anything that sounds off: a mispronounced name, a pause in the wrong place, the wrong word stressed, or a cheerful tone on a sad line.
  6. Fix problems by editing the text. The voice takes its cues from what you write, so the text is your main control. Commas and full stops create pauses. Shorter sentences slow the pace. An exclamation mark adds energy. For a name it gets wrong, spell it the way it sounds ("Shiv-awn" instead of "Siobhan").
  7. Try the voice settings. Most voices have a stability control. Lower stability gives a more expressive, varied performance that can go over the top. Higher stability is steadier and can sound flat. Generate the same paragraph at both ends and hear the difference.
  8. Regenerate only the part that was wrong. Because every reading comes out a little differently, a second try often fixes a single bad line. Then download your file as an MP3.
  9. Optional: clone your own voice. Record one to two minutes of yourself reading naturally in a quiet room, with no music or echo, and upload it to the instant voice cloning tool. ElevenLabs says that going beyond three minutes doesn't help. Only ever do this with your own voice, or with a voice whose owner has clearly agreed.
  10. Check the rules before you publish. Look at what your plan allows for commercial use, and check whether the platform you're publishing to requires you to disclose AI narration.

Common mistakes and misunderstandings

Cloning a voice that isn't yours. Copying a friend, a celebrity, or a narrator you admire without their clear permission breaks ElevenLabs' rules, can break the law, and is simply not yours to take. The legitimate examples above all have one thing in common: consent.

Assuming the free plan covers business use. At the time of writing, the free plan does not include a commercial licence, and anything you publish from it must credit ElevenLabs. The paid plans include commercial rights. If you plan to sell the audio or use it to promote a business, read the current terms and expect to need a paid plan. Terms change, so check them yourself instead of relying on an article, including this one.

Thinking one click makes a finished audiobook. AI narration still needs a human to listen to every minute. Expect trouble with unusual names, abbreviations, numbers and dates, and with words that are spelled the same but said differently, such as "read", "lead", "tear", and "wind". Now and then a voice will also skip or slur a word. Nobody will catch these things except you.

Assuming every shop accepts AI narration. Rules differ between audiobook retailers and they keep changing. At the time of writing, Spotify accepts audiobooks narrated with ElevenLabs through its Findaway Voices service and labels them as digitally narrated. Amazon's ACX service, the usual route into Audible, requires a human narrator, although Amazon offers its own separate "virtual voice" tool to authors who publish through Kindle Direct Publishing. Find out where you can sell before you spend your weekends producing.

Feeding it a poor recording and expecting a good clone. A clone copies everything in your sample, including the echo of your kitchen, the hum of your fridge, and the tinny sound of a phone microphone. A quiet room and a decent microphone matter more than the length of the recording.

Believing it understands your story. It doesn't know that a character is lying or that a line is sarcastic. It guesses the emotion from clues in the text. If a line comes out wrong, give it better clues: change the punctuation, add a phrase like "she whispered", or split a long sentence in two.

Forgetting that your own voice can be copied too. Anyone with a public video or podcast has already published enough audio to be cloned. Criminals use cloned voices in phone scams, often pretending to be a relative who urgently needs money. A simple family safe word, which you agree in person and never post online, defeats this. If a call ever feels wrong, hang up and ring the person back on the number you already have.

Mixing it up with AI music tools. ElevenLabs is best known for the spoken voice. It added a music generator of its own in 2025, but tools like Suno and Udio are built specifically for creating songs and instrumental music.

Related elements