Free Chrome extension · runs on your computer

Listen to any page, in a voice that sounds like a person.

Select text, press Read, and it reads aloud while the sentence lights up on the page. The speech comes from Kokoro, a neural text-to-speech model that runs inside your browser. No account, no subscription, and nothing you read is sent to a server.

Add to Chrome, it's free

53 voices
8 languages
0 bytes of your text uploaded

Try it

This is the real player.

The pill and the settings panel below are the extension's own components, running on this page. The audio was generated with Kokoro ahead of time, so you don't have to download the model to hear it.

1Select a few sentences in the article
2Press the Read button that appears
3Click any sentence to jump, or open settings to change the voice

your-favourite-blog.com/why-i-stopped-skimming

Why I stopped skimming

Most of what I read happens in a browser tab, usually with three other tabs begging for attention. Now I select a paragraph, press Read, and let a calm voice carry me through it while my eyes follow the highlight.

The speech comes from Kokoro, a small text-to-speech model that runs on my own laptop. Nothing I read is uploaded anywhere, and it keeps working on a plane.

It also reads the messy parts the way a person would. A €12.50 ticket on 2024-01-15, a 10GB download, a thank-you to @jane_doe and a link to https://github.com/hakubo/ReadIt all come out sounding natural.

Quotes get a voice of their own.

Reading is to the mind what exercise is to the body.

Try another voice in the settings, change the speed, or click any sentence to jump straight to it.

Keyboard: Space play/pause, ← → previous/next sentence, ↑ ↓ speed.

Features

Made for reading long things.

Says it the way you would

Speech engines tend to read symbols literally. Read it! rewrites them first, so prices, dates and links sound the way a person says them.

Examples of written text and how it is spoken
On the page	What you hear
`€12.50`	“12 euros 50 cents”
`2024-01-15`	“January 15th, 2024” (or “15th January” with a British voice)
`https://www.github.com/a/b`	“github.com”
`@JaneDoe` · `#MachineLearning`	“Jane Doe” · “Machine Learning”
`10GB` · `60 km/h` · `25°C`	“10 gigabytes” · “60 kilometers per hour” · “25 degrees Celsius”
`Henry VIII` · `World War II`	“Henry the Eighth” · “World War 2”

The highlight follows along

The sentence being read is highlighted on the page itself, and the page scrolls to keep it in view. Scroll away and it waits a few seconds before following again. Click any sentence to jump there.

Finds the article

With nothing selected, it reads the main text and skips menus, footers, cookie banners and comment sections. If it guesses wrong on a site, point at the right part once and it remembers.

53 voices, 8 languages

American and British English, Spanish, Brazilian Portuguese, Italian, Hindi, Japanese and Chinese. Text marked as another language switches to a matching voice, and quotes can use a second voice.

Reads what the page means

Image descriptions are read out and code blocks and struck-out text are skipped, each with its own switch in the settings. Keyboard keys and exponents are said properly.

Per-site settings

A different voice for one site, a faster speed for another, extra parts to skip. Your own text rules let you fix how a word is pronounced.

Stays out of the way

A small pill you can drag anywhere or shrink to a play button. Light or dark. Speed from 0.6× to 2×. Turn it off for a site with one click on the toolbar icon.

Why it's different

A good voice usually costs a subscription and your text.

Most readers either use the robotic voices built into your system, or send what you read to a cloud service and charge by the month. Read it! runs a neural voice locally, so you get neither trade-off.

Read it! compared with built-in voices and cloud voice services
	Read it!	Built-in system voices	Cloud voice services
Natural-sounding voice	Yes, a neural model (Kokoro)	Mostly robotic	Yes
Your text stays on your computer	Always	Depends on the voice	No, it's sent to their servers
Price	Free, no limits	Free	Usually monthly or a character quota
Account needed	No	No	Usually
Works offline	Yes, after the first download	Local voices only	No
Prices, dates and links read naturally	Yes (English voices)	No	Varies
Highlight on the page itself	Yes, sentence by sentence	Varies	Often in a separate reader view
Source code you can read	Yes, on GitHub	Varies	Rarely

Limits

What to know before you install.

Running a voice model on your own machine has costs. Here they are, plainly.

A one-time 326 MB download: The model downloads from Hugging Face the first time you press Play, and each voice (about half a megabyte) the first time you use it. After that everything is cached and works offline.
Speed depends on your computer: With a GPU that Chrome can use (WebGPU), speech is generated faster than it's spoken. On older or CPU-only machines the first sentence can take a few seconds, and very long sentences can make playback pause while the next one is made.
Chrome on a desktop or laptop: It needs features of desktop Chrome. There's no mobile version, and other Chromium browsers aren't tested.
English gets the most care: The other seven languages are read well, but the rewriting of prices, dates and units is English only. There are no voices yet for languages such as German, French or Polish.
Some pages are off limits: Chrome doesn't let extensions into its own pages, the Chrome Web Store or the built-in PDF viewer. Text drawn as pixels on a canvas (the Google Docs editor, for example) can't be read either.
It uses memory while it reads: The model stays in memory while it makes speech and for three minutes after, so the next read starts quickly. Then it is unloaded.

Nothing to sign up for, nothing to track.

No account, no analytics, no ads. The only things the extension downloads are the model and voice files, from Hugging Face. What you read is turned into speech on your computer and goes nowhere else. This page doesn't track you either.

Read the privacy policy

FAQ

Questions

Is it really free?

Yes. There's no paid tier and no usage limit, because there's no server to pay for: your computer does the work. If you'd like to support it, there's a coffee button in the player.

How do I read a page without selecting anything?

Press Alt+R, or press Play on the pill with nothing selected. It finds the main text and starts from the top.

Why is the download so big?

That's the voice model itself. Smaller, compressed versions exist, but they sounded noticeably worse, so the extension uses the full one. You download it once.

Which voice should I pick?

Heart (American English) is the default and a good start. Open the settings, press the play button next to any voice to hear it, and pick the one you like listening to for an hour.

Does it work offline?

Yes, once the model and the voice you use have been downloaded.

Can websites tell I have it installed?

A site could spot one small file it makes available, the audio player that lets it play sound on sites with strict security rules. That's true of most extensions. Read it! never sends a site anything, and it asks for as few permissions as it can.

Give your eyes a break.

Add Read it! to Chrome