181 / 20,000

2 sentences. Start a line with a voice name in brackets, like [George] Hello then [Emma] Hi, to give each speaker their own voice.

A short preview needs the voice model (about 90 MB, downloaded once) if it isn't on this device yet.

1.00×
Hear all the voices

American female

American male

British female

British male

0.6 s

Also used when the speaker changes. Applies to downloads and to sentences not yet played.

Pronunciation rules

Say a name or acronym your own way: type the word as it appears, then how it should sound, for example Nguyen as win, or SQL as sequel. Whole words only. Rules are kept in this browser.

The voice model (about 90 MB) downloads once on first use and is then cached by your browser. Your text is never sent anywhere — the speech is made on your device.

  • 100% Free
  • No Sign-Up
  • No Credit Card
  • No Free Trial
  • No Pro Tier
  • No Watermarks

Nothing is uploaded. Just open it and start.

Turn text into natural speech

Free, no sign-up, no API key, and no per-character fees. Paste your text, pick a voice, and hear lifelike speech start within seconds. Fix pronunciations, script a dialogue, then download MP3, WAV or subtitles. It runs on Kokoro, an open neural voice model that works entirely inside your browser — so your words never leave your device.

Plays as it renders

Hear the first sentence in seconds while the rest is made, with a read-along that highlights each sentence and jumps where you click.

Many voices

Hear every voice first, then give each speaker their own with a [Name] script, and set how long the pauses are.

MP3, WAV and subtitles

Download audio levelled to speech volume, plus SRT or WebVTT subtitles that match it.

Private & free

Runs on your device in a background worker: no sign-up, no key, no per-character fees, nothing uploaded. Up to 20,000 characters per clip.

Frequently asked questions

Is it really free?
Yes. The voice runs on your own device, so there is nobody paying per character and nothing to meter: no account, no API key, no watermark and no per-character fees. One clip can hold up to 20,000 characters (roughly 20 minutes of speech), and you can make as many clips as you like. Most free text-to-speech sites are a front end for a paid cloud API, which is why they limit you to a few hundred characters or add a watermark.
Which voice does it use?
Kokoro, an open-weight 82-million-parameter speech model released under the Apache 2.0 licence. It is small enough to run in a browser tab yet rates well against far larger models, which is what makes free and private possible at the same time. It offers a range of American and British voices, female and male, and you can hear each one before you make anything.
Does it start talking straight away?
As soon as the first sentence is ready. The text is made one sentence at a time and each sentence plays while the next is being made, so a long text does not make you wait for the whole thing. If your device is slower than real time the speech pauses between sentences until the next one lands. Untick the option to start playing early if you would rather have the whole clip first. The full clip is kept for download either way.
What does Stop do?
The voice runs in a background worker, so the page stays responsive and Stop ends the work at once, including during the first download. Leaving the page does the same. If your browser cannot start a worker, the voice runs on the page itself and the tool tells you so; Stop then takes effect after the sentence being made. After Stop the sentences already made are kept, marked as a partial clip, and you can make the rest.
What is the read-along?
The text you rendered is shown sentence by sentence and the sentence being spoken is highlighted. Click a sentence, or move to it with the arrow keys and press Enter, to play from there. You can also pick a sentence and remake just that one, for example after respelling a word the way it should sound.
Can I fix a name or acronym it gets wrong?
Yes, with pronunciation rules: type the word and how it should be said (for example SQL as sequel). Rules match whole words, are kept in this browser, and apply to every clip. Numbers, money, percentages, ordinals, years and times are also read as words by default (for example $4.50 as four dollars fifty cents), which sounds right and keeps long numbers within what the model can take; you can switch that off.
Can I make a dialogue with several voices?
Yes. Start a line with a voice name in brackets, such as [George] Hello and then [Emma] Hi there, and that voice speaks it and every line after it until the next tag. Use a voice's name from the list (or its id). A bracket that is not a voice name is read out as ordinary text, and the page points it out. The pause between paragraphs, and between speakers, is adjustable.
Does it support Hindi or other languages?
Not yet. It speaks English, in American and British voices. The Kokoro model itself has Hindi, Spanish, French, Italian, Portuguese, Japanese and Chinese voices, but pronouncing them needs a pronunciation engine for each writing system, and none of those run in a browser today. Paste another script and the tool will say so rather than reading it as nonsense.
Is my text uploaded anywhere?
No. Your text stays in your browser and the speech is made on your device. The one thing that comes over the network is the voice itself: the speech library from a public CDN and the voice model (about 90 MB) from the Hugging Face hub, downloaded once on first use and then cached by your browser. Your voice, speed, options and pronunciation rules are remembered in this browser only.
Why is the first run slow?
Only the first one is. The model has to download (about 90 MB) and then warm up. Afterwards your browser has it cached. Speed depends on your device: a recent laptop or phone renders several times faster than real time. On browsers that support it there is an experimental option to use the graphics chip; it uses the same files, and falls back to the processor by itself if the chip cannot run the voice.
What can I download?
MP3 or WAV, encoded on your device, plus SRT and WebVTT subtitles built from the clip's own timings (one cue per sentence, with the speaker's name added in a multi-voice script). File names come from the first words of your text. You can level the volume (to about −16 LUFS, an approximate measurement) so that quiet and loud voices match. A clip that was stopped early is marked partial in the page and in the file name.
Can I use the audio commercially?
The Kokoro model is Apache 2.0 licensed, which permits commercial use of what it produces, and we add no restriction of our own, so the audio is yours. If it matters for your project, check Kokoro's own model card for the current terms rather than taking a web page's word for it.
How is this different from my browser's built-in speech?
The Web Speech API uses whatever voices your operating system ships, which vary by device and mostly sound robotic, and on many browsers it cannot be recorded or saved at all. Kokoro is a neural model that sounds markedly more natural, sounds the same on every device, and gives you a file you can download.

Spotted a bug or have a suggestion?

Found something broken, have an idea, or just want to say thanks about any of our tools? Every message reaches a real person.