Free speech to text, right in your browser.

How voice typing online works without uploading your voice

When the page loads, it downloads a speech recognition model called Whisper into your browser. It is the base size, quantised to int8, and comes to about 80MB. A progress bar shows the download. The model travels to you instead of your voice travelling to a server.

After that the tool is push to talk. You tap the mic, your browser asks for microphone permission, and a level meter shows it is hearing you. Tap again and the recording goes straight to the model running in the same tab. The audio is handled in 30 second windows with a 5 second overlap, so a long stretch of talking is not chopped up mid sentence.

Where your browser supports WebGPU, the model runs on the graphics card and is several times faster. Where it does not, it falls back to WebAssembly on the CPU, which is slower but still works. Your browser caches the model files after the first run, so later visits start immediately. A word count and a rough words per minute figure sit above the text box.

Voice typing with no upload, no signup and no account

Most free voice typing sites take one of two shapes. Some record your audio and post it to their servers for transcription. Others call the speech recognition built into Chrome or Edge, which sends the audio to the browser vendor's service. Either way your voice leaves your machine and you are trusting a privacy policy you did not write.

This page has a different shape. There is no upload, no account and no transcription server. The model file is a static download, and the transcription runs in the tab in front of you. You can check that yourself. Load the page, wait for the model to arrive, then turn off your wifi and keep dictating. It carries on working, because the recognition was never using the network.

That structure has practical effects too. Nothing is metered, so there is no minutes cap, no daily limit and no wall asking you to sign up halfway through a recording. It also means we cannot read, store or lose your transcript. If you close the tab, the text is gone, so copy or download anything you want to keep.

How to use the voice typing tool, step by step

Wait for the model to finish downloading. The mic button stays inactive until it is ready, because there is nothing to transcribe into before then. On a normal connection this takes a few seconds, and it only happens on your first visit.

Tap the mic. The first time, your browser will ask for permission to use the microphone. Allow it. Watch the level meter move as you speak, which is the quickest way to confirm the right input device is selected. If the meter is flat, change your input in your system sound settings and reload.

Talk at your normal pace. There is no need to slow down or over enunciate, and shouting makes results worse rather than better. Tap the mic again when you finish a thought. The text appears in the box below.

You can keep going. Each new recording is added to the end of what is already there, so you can dictate in short bursts. The box is a normal editable field, so fix anything you like, then copy it or download it as a .txt file.

How accurate is it, and which languages and accents does it handle?

The model is multilingual and detects the language automatically, so you can start speaking without picking anything from a menu. In practice that covers most major languages, and it is noticeably stronger on the ones with the most training data, English chief among them.

Be realistic about the size of the model. This is Whisper base, chosen because an 80MB download is something a visitor will actually wait for. It is good at ordinary connected speech and it holds up on a range of accents, but it is not the largest model available and it will make more mistakes than one that is ten times the size. Proper nouns, product names, technical jargon and people's names are where it slips most often.

A quiet room and a decent microphone help more than anything else you can change. Laptop microphones pick up keyboard noise and room echo, and both make the text worse. If you dictate a lot, a cheap headset is the single biggest upgrade available to you.

What people use free voice typing for

The most common use is a first draft. Talking is faster than typing for most people, and a rough spoken draft that you then edit tends to beat a blank page. Essays, blog posts, cover letters, long emails, journal entries and meeting notes all work well this way.

It is also a straightforward accessibility tool. If typing is painful, if you have an injury, or if a keyboard is simply slow going for you, dictating into a browser tab needs no install, no admin rights and no purchase. That matters on a locked down work laptop or a shared library machine where you cannot install software.

Students use it for notes and for getting long quotes down quickly. People writing in a second language often find it easier to say a sentence than to type it. And it is useful as a scratchpad: capture the idea while you have it, tidy it later.

If what you have is an existing recording rather than a live voice, that is a different job. Use the audio to text tool for that instead.

Voice typing offline, without an internet connection

Once the model is in your browser cache, this page keeps working with no connection. The transcription was never using the network in the first place, so a plane, a train or a dead hotel wifi does not stop it. You do need the page itself to load, so open it before you lose signal.

Two things send you back to the download. Clearing your browser storage removes the cached model, and a private or incognito window usually starts with an empty cache, so it fetches the file again. Different browsers keep separate caches too, so switching from one to another means one more download.

That first download is the only moment this tool needs the internet. Everything else, the recording, the recognition and the text, happens on your own hardware. It is worth saying plainly because it is unusual. Most free dictation sites stop working the moment you go offline, because the recognition is happening somewhere else.

The limits of browser voice typing, and what the desktop app adds

This tool types into one box on one web page. That is the honest boundary. It cannot put text into your editor, your email client, Slack or a document, so you copy the result out by hand every time. If you want to dictate into something else, this is a two step job.

The text also arrives after you stop, not word by word as you speak, because the recording is transcribed in one pass when you tap the mic again. And there is no cleanup pass. Whatever you said is what you get, including the ums, the false starts and the sentence you abandoned halfway through.

The Yaps desktop app is the same on-device idea without those limits. You hold one hotkey, speak into whatever app you are already in, and the text lands at your cursor. It cleans up filler words and punctuation as it goes, and it works offline once installed. There is an Android keyboard too, which is a much better fit for phone dictation than a browser tab is.