Finished reads appear here, newest first, ready to play.
Paste your words, pick a voice from the ElevenLabs library and get a natural read back as an MP3 in seconds: narration, characters, emotion and 32 languages.
Try it freeBrowse the ElevenLabs library by accent, age and use, and switch voices on every run. A calm narrator for an audiobook, an energetic host for an ad, a villain for a game.
On ElevenLabs v3, write [whispers], [laughs] or [excited] into the text and the voice performs it instead of reading the word aloud.
Turbo and Flash v2.5 cover 32 languages and Multilingual v2 covers 29. Paste the text in any of them, or mix several in one script, without setting a language anywhere.
Jonathan Livingston
Multilingual v2 · calm narrator · 0:09
Brielle
Multilingual v2 · natural, conversational · 0:09
Every city has a sound before it wakes up. In Lisbon it is the first tram climbing the hill, the cafés rolling up their shutters, and the river turning from grey to gold.
Multilingual v2 is the model ElevenLabs points to where lifelike speech matters. Hear the same paragraph in two voices below: the words are identical, the read is not. Pick the voice that suits the story.
Jessica Gallagher
ElevenLabs v3 · with audio tags · 0:10
Brielle
ElevenLabs v3 · with audio tags · 0:08
[whispers] Okay, don't tell anyone, but I found the old key. [excited] It still works! [laughs] Nine years in a drawer and it still works.
Switch to ElevenLabs v3 and put stage directions in square brackets. The voice whispers, gets excited and laughs where you told it to, and the tags themselves are never spoken.
Xavier
Turbo v2.5 · game announcer · 0:14
Britney
Turbo v2.5 · calm villain · 0:10
Announcer: "Checkpoint reached, recruit. Reload, catch your breath, and stay behind the barricade. The next wave comes in thirty seconds." Villain: "You knew. All this time, you knew, and you let me walk in there anyway. Don't. Don't say my name like that."
The library has voices built for performance: announcers, villains, cartoon characters. Cast one per line and you have a scene read without booking a studio.
Jessica Gallagher
Multilingual v2 · five languages · 0:10
Welcome to Lisbon. Bienvenidos a Lisboa. Bienvenue à Lisbonne. Willkommen in Lissabon. リスボンへようこそ。
Write each line in its own language and the same voice reads them all, keeping its character from one language to the next. Useful for travel videos, language lessons and product demos for several markets.
Up to 1,000 characters a run, about a minute of speech. For longer scripts, generate one section at a time with the same voice.
Open the voice library and filter by accent, age or use, or start from one of the examples, which come with a voice already cast.
Turbo v2.5 for quick drafts at 15 credits, Multilingual v2 for the most natural narration at 30, and v3 at 30 when you want emotion tags.
The read is ready in seconds. Play it, then download the MP3, or change a word and run it again.
Which model to draft on and which to finish on, how emotion tags work, and why a voice carries across languages, in ElevenLabs' own documentation.
Draft on a fast model at 15 credits, then make the final read on Multilingual v2 at 30. ElevenLabs itself says v2 is slower and costs more than its fast models, but sounds more lifelike.
“While it has a higher latency & cost per character than Flash models, it delivers superior quality for projects where lifelike speech is important.”On Multilingual v2 you pick a voice for how it sounds and keep it across languages: ElevenLabs says the voice keeps its character and accent in every language the model supports.
“The model delivers consistent voice quality and personality across all supported languages while maintaining the speaker’s unique characteristics and accent.”On v3, bracketed tags such as [whispers], [laughs] and [excited] are directions, not words: ElevenLabs lists them as the tags that control how a line is delivered and what emotion it carries.
“These tags control vocal delivery and emotional expression”Software that reads written text aloud. AI text to speech, like the ElevenLabs models on this page, uses voices trained on real recordings, so it pauses, stresses words and changes pace the way a person reading would.
Paste your text, pick a voice and select Generate. The read is ready in a few seconds; play it on the page and download it as an MP3.
Yes, to start. Sign up for a free account and your first reads on Turbo v2.5 cost nothing. After that a run is 15 credits on Turbo or Flash v2.5 and 30 on Multilingual v2 or v3, or start the 3-day free trial for 200 credits. Starting the trial takes a card but costs nothing today; we email you before it ends, and if you do not cancel, the plan you picked renews at its normal price.
No. Generating a read takes a free account, which is also where your reads are kept to play and download later.
Pick Multilingual v2 for the most natural read, then choose a voice labelled for narration or conversation in the library. Neutral voices stay the steadiest across long reads and languages; character voices are made to perform.
Yes, on ElevenLabs v3. Write directions in square brackets, such as [whispers], [laughs], [excited] or [sighs], and the voice performs them instead of reading the words. Punctuation helps on every model: an exclamation mark lifts the line, an ellipsis slows it.
Yes. The voice library includes announcers, villains, cartoon and game characters alongside narrators. Filter by use to find them. Voices of real celebrities are not offered.
Turbo v2.5 and Flash v2.5 read 32 languages, Multilingual v2 reads 29 and v3 more than 70. You do not set the language: the model reads what you paste, and one script can switch between languages.
1,000 characters per run, which is about a minute of speech. For a longer script, generate it in sections with the same voice and model, then join the files in an audio or video editor.
Yes. Every read downloads as an MP3. On a free account the file carries a short, quiet tone every ten seconds; paid plans and the free trial download the clean file.
No. Voices come from the ElevenLabs library; this page does not clone a voice from a recording.
Yes. Speech you generate is private to your account by default and does not appear in Explore.
