Skip to content

Text to Speech: Natural English Voices

Turn text into natural speech with the Kokoro AI voices, in your browser. 28 American and British English voices, any speed. Download as WAV.

Private · on device. Runs in your browser: your file is not uploaded. No sign-up.

Text

Paste or type English text. Long text is read sentence by sentence.

0 characters

Checking for the AI model on this device…

Natural voices that run in your browser

This tool uses Kokoro, a small open text-to-speech model (82 million parameters, Apache-2.0 licence) whose voices sound far more natural than the robotic voices built into most devices. The model is downloaded once from Hugging Face and kept by your browser, and the speech is created on your own device, so nothing you type is sent anywhere.

How to use it

  1. Paste or type your text. Paragraphs and sentences are kept, with a short pause between parts.
  2. Pick a voice and a speed, and press "Preview this voice" to hear it.
  3. Press "Create speech". Listen in the page, then download the WAV file.

Tips for better speech

  • Write numbers, dates and abbreviations the way they should be said when it matters ("5 kg" as "five kilograms").
  • Use punctuation: commas and full stops give the voice its pauses.
  • Slower speeds suit learners and announcements; 1.1× to 1.25× suits narration.

What it does not do

It speaks English only, and it saves WAV only, not MP3. It cannot clone a particular person's voice. To turn a PDF into speech, copy its text with PDF to text first; for the reverse, from a recording to text, use audio to text.

The software behind this tool

Speech is made by the Kokoro model (Apache-2.0) through kokoro-js, which turns words into sounds with eSpeak NG, compiled to run in your browser. eSpeak NG is free software under the GNU General Public License version 3; its source code is at github.com/espeak-ng/espeak-ng and the browser build at github.com/xenova/phonemizer.js. The source of this page's speech worker is available on request through the contact page.

Frequently asked questions

Does it speak Hindi or other Indian languages?

No. The Kokoro voices in this tool speak English only, with American or British accents. Text in Devanagari or another Indian script is skipped or read wrongly, and Hindi written in English letters is read with an English accent.

Is my text sent to a server?

No. The voice model is downloaded to your browser once (about 92 MB) and the speech is created on your device. Each voice adds a small file of about half a megabyte the first time you use it.

Can I download MP3?

Not from this tool: it saves WAV, which plays on every phone, computer and video editor. WAV files are larger (about 2.9 MB a minute), so convert to MP3 elsewhere if size matters.

How long a text can it read?

Up to 20,000 characters at a time, roughly 25 minutes of speech. Long text is read sentence by sentence and the progress bar shows how far it has got. A computer creates speech several times faster than it plays; a phone is slower.

Can I use the audio in a video or commercially?

The Kokoro model is released under the Apache-2.0 licence, which allows commercial use, and this tool adds no restrictions. You are responsible for the text you have read out.

Which voice sounds best?

Heart and Bella (American female) are the most natural, followed by Nicole, Emma (British female) and Michael (American male). Use "Preview this voice" to hear one before creating the full audio.