Free text and speech tools that run in your browser

Which tool do you actually need?

These three tools move words between three states: locked inside a picture, locked inside audio, or sitting on the page as text. Working out which direction you are going in answers the question immediately.

The search box takes plain words. Typing proofread finds the reader, typing ocr finds the picture tool, because it matches what each one does rather than only what it is called.

Reading words out of a picture

A photograph of a page contains no words. It contains pixels arranged in shapes that your eye assembles into letters. Optical character recognition turns those shapes back into characters, guessing at each one and reporting how confident it felt.

Accuracy is decided almost entirely by the picture, not by the engine. A screenshot of text on a screen is close to perfect, because nothing is blurred or tilted. A phone photo taken at an angle in poor light is a different story.

Four things help more than any setting:

• Make the letters bigger. Zoom the page in before you screenshot it. Recognition needs enough pixels per character to tell a lowercase L from a one.
• Crop tightly. Cut away the desk, the margins and anything decorative.
• Straighten the page. Even a small tilt costs accuracy, and most gallery apps fix it in seconds.
• Shoot parallel to the paper in even light, with no shadow falling across it.

Names, acronyms and technical jargon are the words that come out wrong most often, because the engine has rarely seen them. Those are the parts to check before you use the text anywhere that matters.

Turning a recording into text

Transcribing an hour of audio by hand takes most people four to six hours of stopping, rewinding and typing. Getting a rough transcript in minutes and spending twenty of them correcting it is a completely different afternoon.

What you get is a first draft, not a finished document. Clear speech from one person with a close microphone produces something you barely need to touch. A meeting room with three people talking over each other produces something you will sit with. Crosstalk is genuinely hard for any transcriber, human or otherwise.

The recording itself decides most of it. Get the microphone close, choose a room with soft furnishings, turn off anything humming, and ask people not to speak over each other. Having everyone say their name at the start gives you an anchor when you label the speakers afterwards, since no automatic transcriber separates voices reliably.

The privacy difference nobody else mentions

This is the part worth reading carefully, because the tools on this page are not all the same in this respect and most sites blur it.

The file based tools are genuinely local. The recognition engine or the speech model downloads to your browser once, then everything happens in the tab. Your scan or your recording never goes anywhere. You can watch the network panel and confirm it.

Live dictation is not. It uses the speech recognition built into your browser, and in Chrome and Edge that works by streaming your audio to Google to be turned into words. It is fast, needs no download, and is excellent for jotting a note. It is a poor choice for a confidential conversation.

Plenty of sites offer a microphone button and describe the whole page as completely private. That is not accurate, and we would rather say so on the button than let you assume. For anything sensitive, record it and use the file route.

Having text read to you

Most people meet text to speech through accessibility software and assume that is the whole story. It is more useful than that.

Proofreading, which is the use people underrate

Your eye reads what it expects to be there, especially in your own writing, which you have already read eleven times. That is why the missing word survives four rounds of checking and then leaps out the moment somebody else opens the document.

Your ear has no such loyalty. Listening to your own writing catches repeated words, sentences that run out of air, four paragraphs that all begin with However, and the sentence that only made sense inside your head. Slow it down slightly and follow the highlighted words while you listen.

Studying and getting through reading

Listening while reading engages two routes into the same material, and for many people that makes it stick better. It also stops the drift where your eyes travel down the page while your mind goes elsewhere, because the voice keeps moving and takes your attention with it.

Everyday use

Long articles while cooking or walking. Resting your eyes after a day of screens. Making written content usable for someone who is blind or has dyslexia. None of these require a special reason.

Why the first run is slow

Each of these tools needs something downloaded before it can start. A language file for reading text, a speech model for transcription. Anywhere from a couple of megabytes to about eighty depending on which tool and which option.

That happens once. Your browser caches it, so every run afterwards begins immediately, and it is also why these keep working with your connection switched off once you have used them. If the first attempt feels slow, that is the download rather than the work.

Text to speech is the exception. It uses voices already installed on your device, so there is nothing to fetch and it starts instantly.

The honest limits

• Handwriting cannot be read. The engine learned printed characters. Neat block capitals occasionally survive, ordinary joined up writing does not, and no setting changes it.
• PDFs are not pictures. A PDF page is drawing instructions until something renders it. If yours is a scan, screenshot the page and read that. If it already contains real text, select and copy it in any reader instead.
• Layout is flattened. Columns, tables and text boxes come out as ordinary lines. Crop and read one column at a time for anything complicated.
• Transcription here is English only. The models small enough to run comfortably in a browser are the English ones.
• Names always need checking. In both reading and transcription, a misheard surname in something published is a bigger problem than a hundred small typos.

Using the three together

They chain in ways that are more useful than any one of them alone.

A photographed page becomes text, and that text can then be read aloud, which turns a book chapter into something you can listen to on a walk. A recorded lecture becomes a transcript you can search for the one thing you needed. A draft you wrote gets read back to you before you send it.

If your source is a document rather than a picture, the PDF tools handle turning pages into images first. If a screenshot is too large or the wrong format for the reader, the image tools sort that out. And if what you are assembling is an application rather than a file, the career tools cover resumes and letters.

What stays on your device, and why it matters

Think about what actually goes through these tools. Payslips and bank letters photographed for a form. Medical results. Passport pages for a visa application. Interviews with sources who spoke on condition of anonymity. Internal meetings about restructuring. Drafts nobody has seen.

This is among the most sensitive material a person handles, and the ordinary route is to upload all of it to a company you know nothing about. Several transcription services have used customer recordings to train their models. That may be acceptable for a holiday video and completely unacceptable for a confidential interview.

Here the model comes down to you and your file stays put. Open the network panel in your developer tools before you start and watch it happen if you would rather see it than be told. There is no account, no daily allowance, no watermark and nothing to unsubscribe from afterwards, because there was never anything stored to begin with.