Text to Speech Guide

Open the app — Text to Speech

Have you ever read your own writing out loud?
Sentences that looked fine on screen suddenly snag when they leave your mouth.
A clause that runs too long, a word you used twice, a phrase that doesn’t quite land.

But reading aloud every time is a chore.
In an office you feel watched, and with a long piece your throat gives out.
It would be easier if someone else read it — though signing up for a TTS service just for that feels like too much.

This tool uses the voices your browser already has.
No sign-up, nothing to install, no API key. Paste, press, and it reads.

How to use it

The basic flow

1. Put in your text. Type it or paste it.
2. Pick a voice. One is chosen for you when the page opens, so you can leave it alone.
3. Press the round button at the bottom.
4. As it reads, the current sentence is shown on the right.

There is only one button.
When stopped it is a mint play shape; while reading it turns into a coral stop shape.
Both the colour and the shape change, so you can never mistake which state you are in.

Once you stop and press again, it reads from the beginning.
To hear a specific part, use the trick below.

Click a sentence to start there

Click any sentence on the right and reading starts from that sentence.

This helps when you want to hear one passage again in a long piece.
No need to play from the top and wait through everything before it.

If you work by keyboard, tab between sentences and press Enter or Space to pick one.

Three ways to get text in

Typing is the default, and there are two more.

• The [Sample] button — drops in an example. It mixes periods and question marks, a sentence inside quotation marks, numbers and a Latin-script name, so you can see at a glance how the tool splits sentences and how they sound.
• Load a file — opens txt, md or csv files.

If a file exceeds the character limit, it is not loaded and you are told so.
Trimming it silently would hide the fact that the rest vanished, so it seemed better to say so and let you choose what to keep.

Picking a voice

The list shows the voices this device can use.
It keeps to your interface language and English, because operating systems ship voices for dozens of languages and listing them all buries the one you actually want.

Watch for the ☁ mark next to a name.

When you first open the page, a ☁ voice is chosen.
The gap is larger than you would expect. Device voices articulate clearly but the intonation stays flat, while a voice synthesized by a neural network on a server comes close to a person reading.

Outside Chrome-based browsers there are no ☁ voices in the list, and a device voice is chosen instead.

Speed, pitch and volume

Three things are adjustable.

There is a reason none of them open all the way.

The spec allows speed up to 10, but past 2 some browsers leave the synthesis engine stuck.
When that happens the signal that a sentence has finished never arrives, so it cannot move to the next one and playback simply stops there.

Pitch runs from 0 to 2, but at either extreme it stops sounding like a human voice.
Volume can reach 0, where nothing plays at all — easy to mistake for a broken app, so the bottom is closed off.

The ↺ button beside each slider restores the default.

Change a value mid-reading and the current sentence restarts.
The browser does not apply new values to a sentence already in progress, so re-reading it is the only way to let you hear the change right away.

Chimes

You can pick a short sound that marks the start and the end of reading.

The start rises and the end falls.
That way you can tell one from the other by ear without looking at the screen.

Preview them with the [Start] and [End] buttons. They follow the volume you set above.

Each sound fades out over two to three seconds.
A short beep feels like being hurried along, but the same sound with a tail to it lands gently.

Once the start chime fades, reading begins after a one-second pause.
Speech that starts before the sound clears trips over it and turns messy. The end works the same way — the chime sounds one second after the last sentence finishes.

Nothing is downloaded for this: the sounds are generated on the spot rather than stored as audio files.

Good to know

Commas are what make it sound natural

With the same text, where you place your commas changes the reading a great deal.

A synthesis engine pauses at a comma and shapes the intonation around it.
A long sentence with no commas gets pushed out in one breath, and that is what sounds unnatural.

If a passage sounds off when you hear it, try putting a comma where you would take a breath.
It improves things noticeably without rewriting the sentence.

The limit is 600 characters

This was built as a tool for checking short pieces, not for having long documents read to you.
Think of proofing a draft by ear, or comparing how a few sentences land.

The input box stops accepting text at 600 characters. The count reads like 182 / 600 and turns red when it fills up.

☁ voices send your text to a server

Voices marked ☁ are synthesized on the browser maker’s servers.
That means the text you want read is sent there.

For anything sensitive, choose a voice without the mark. Those run inside your device, so nothing leaves it, and they work with the connection down.

Traffic is not a concern.
Even at the full 600 characters, the audio coming back stays under 0.5 MB, and the text going out is well under 1 KB.

Do note that re-reading the same passage fetches it again. If you plan to listen repeatedly, a device voice costs no traffic and starts faster.

How sentences are split

Rather than handing over the whole text, this tool cuts it into sentences and reads them one at a time.
It has to know which sentence is playing in order to show it on screen, and this also sidesteps a long-standing problem where Chrome cuts out around fifteen seconds into a long utterance.

The rules are these.

• It breaks at periods, question marks, exclamation marks and ellipses.
• Closing quotes and brackets stay with the sentence before them. He said "hello" and left. reads as one sentence.
• Runs like ?! or ... are kept together.
• Line breaks count as sentence boundaries.
• No break when a letter follows the mark directly. 3.14 and index.html are never cut in the middle.
• If a sentence still runs past 180 characters, it splits at a comma or a space.

Why the list differs between devices

The voices come from your operating system and browser, not from this tool.
So the same page shows a different list on each computer.

• macOS — Siri voices and the enhanced voices you download in System Settings are not exposed to browsers. Apple keeps them closed off, and there is nothing this tool can do about it.
• Windows — often only one Korean voice, Heami, is installed on the device. On Chrome the ☁ voices appear alongside it.
• Safari and Firefox — no ☁ voices, so only device voices show.

A few entries are left out on purpose.
The novelty voices bundled with macOS — the ones that sing or mumble — do not read text, so they are filtered out, and when the same name appears twice only one is kept.

Install a new voice and it shows up in the list without a refresh.

It remembers where you left off

Your text and your voice, speed, pitch, volume and chime settings are saved in the browser.
Close the tab, open it again, and everything is as you left it.

This is stored only inside this browser and never uploaded. It does not carry across to your other devices.

If you hear nothing

• A ☁ voice with no connection — switch to a voice without the mark.
• iPhone or iPad — the first playback has to happen in the moment you tap the screen. Try pressing once more.
• Volume sitting at 10% — the ↺ beside the slider puts it back.

Tools that pair well with this

For the opposite direction, see Extract Text from Audio Files.
It turns recordings into text, and there the model runs inside the browser so the audio never leaves the page.

If the text you want read has broken Hangul in it, clean it up first with Broken Hangul Fixer.

Open the app — Text to Speech