How to Use Text to Speech in Your Browser (and What It Cannot Do)
Every modern browser can read text out loud, and almost nobody uses the feature. It is built in, it needs no account, and it works the moment a page loads. It is also surrounded by misconceptions — mostly about what it can produce, and where your text goes.
Who is actually speaking
When a browser speaks, the audio comes from a voice engine, and that engine can live in one of two places.
On your device. Windows, macOS, iOS, Android, and Linux all ship speech engines. These work with no network connection at all. They are the voices you find in your operating system’s accessibility settings.
In the browser vendor’s cloud. Chrome, in particular, also offers voices synthesised remotely — they appear in the same voice list as the local ones, usually named after the vendor. These need a connection, and the text is sent away to be spoken.
The practical consequence: if the text is sensitive, pick a voice that is installed locally. Your operating system’s speech settings will tell you which ones are.
Why this is different from a “free text to speech” website
Search for text-to-speech and you will find dozens of sites offering it, most of them wanting an account, a credit card, or both. The reason is simple: they run a speech model on a server, and a server costs money.
That model has a cost for you, too. Every word you paste is transmitted to that server, processed there, and — depending on the site — potentially logged. People paste all sorts of things into text-to-speech boxes: draft emails, internal documents, medical notes, messages they are trying to phrase carefully.
A browser voice sidesteps the entire question. Nothing is transmitted, because the synthesis happens where the text already is. The trade-off is voice quality: an operating-system voice is functional rather than lifelike, and the newest cloud voices sound noticeably better. For listening back to a draft, functional is usually enough.
The thing it cannot do
This is the misconception worth clearing up first, because it wastes a lot of people’s time.
A browser cannot export text to an audio file. There is no download button, and there is no setting that adds one. The speech synthesis API is designed for accessibility — reading a page aloud — and it deliberately does not expose the audio stream to the page. A website cannot record what it is saying.
So if you need an MP3 or WAV — for a video voiceover, a podcast, an audiobook, a phone system prompt — browser text-to-speech is the wrong tool, and no amount of hunting for a hidden export button will change that. You need either a dedicated recording setup or a service that generates a file as its actual product.
What the browser is genuinely good at is listening: proofreading a draft by ear, hearing how a sentence lands, checking the pacing of something you wrote, or listening to a long article while doing something else.
Getting a good result
Two controls do most of the work.
Speed. The default is 1×, and for most content it is too slow once you are used to it. 1.25× to 1.5× is where most people settle when reading back their own writing. Go slower — 0.8× or below — when you are checking individual word choices or listening in a second language.
Pitch. Leave it alone unless you have a reason. Lowering pitch slightly can make a thin voice easier to listen to for long stretches; raising it usually just sounds worse. It is not a gender or character control, despite how it is often described.
Two smaller things worth knowing:
Long text is read as one continuous stream. There is no per-paragraph control, and pausing loses your place in some browsers. For anything over a few hundred words, splitting it into chunks and playing them one at a time is more reliable than fighting the player.
The voice list depends on the machine, not the site. The same page will offer different voices on Windows and on a phone, because the list is whatever the operating system has installed. If a voice you expect is missing, it is not installed — that is a system setting, not a browser one.
When not to use it
If you need audio you can keep, share, or edit — this is not it.
If you need broadcast-quality narration — this is not it either. Operating-system voices are intelligible and clearly synthetic. They are fine for proofreading and accessibility, and wrong for anything an audience will hear as a finished product.
If the text is confidential and you are unsure which voice is being used — check the voice list first. A locally installed voice keeps everything on your machine; a vendor cloud voice does not.
Frequently asked questions
Can I download the audio as an MP3? No. Browser speech synthesis does not expose the audio stream, so no page can record or save it. You need a dedicated tool or service for that.
Does browser text-to-speech need an internet connection? Not if the voice is installed on your device. Voices provided by the browser vendor’s cloud do require a connection.
Is my text uploaded anywhere? With a locally installed voice, no — synthesis happens on your device. With a vendor cloud voice, the text is sent to that vendor to be spoken.
Why are the available voices different on my phone? Because the list comes from the operating system, and each device has a different set installed. Add voices in your system’s speech or accessibility settings.
Which languages can it read? Any language you have a voice installed for — commonly English, Mandarin, Spanish, French, German, and Japanese. Mixed-language text is usually read with the wrong pronunciation for one of the languages.
Is it accurate enough for proofreading? Yes, and it catches awkward phrasing, repeated words, and sentences that run too long better than reading silently does. Your ear notices what your eye skips.