Dictation: speak and get text

Quick dictation uses your browser’s speech service, and says so. Recording on this device keeps the audio here.

Quick dictation

Spoken language

Microphone off

Quick dictation uses the speech recognition built into your browser. In Chrome and Edge the browser sends your microphone audio to its vendor’s servers. For audio that must stay on this device, use “Record on this device” below.

Record on this device

Records your microphone here and transcribes the recording with the local speech model when you stop. The audio is not sent anywhere.

First use downloads the 44 MB speech model from huggingface.co and keeps it in this browser.

Dictated text

0 words · 0 characters · about 0.0 min spoken

Tidy the text

Remove filler words (um, uh, like, you know…)Fix spacing and doubled punctuationCapitalise the start of sentences

Where your voice goes, browser by browser

Live dictation in a web page is not done by the page. It is done by the browser, through an interface called the Web Speech API, and each browser decides how to recognise speech. This site receives only the resulting text, in your browser, and does not send it anywhere.

Browser	Quick dictation	Record on this device
Chrome, Edge	Works. The browser sends microphone audio to its vendor’s speech servers.	Works; verified by us.
Safari	Partly supported. Recognition is handled by Apple.	Expected to work; not verified by us yet.
Firefox	Not available: Firefox does not ship speech recognition.	Expected to work; not verified by us yet.

Which one to use

Quick dictation shows words as you speak, which suits notes, drafts and messages where seeing the text grow helps you keep your thread. It needs a connection.

Record on this device is for anything you would not say to a third party: client details, health, legal or financial matters, unpublished work. You speak, press stop, and the local model writes the text. The price is that the text comes at the end, and that the first use downloads the speech model.

Have a recording already? Transcribe the file instead: you get timestamps and an editor tied to the waveform.

Dictating well

Speak in full sentences and pause between them; recognisers use the pause to decide where a sentence ends.
Pick the language before you start. Changing it restarts the recogniser.
Keep the microphone close and steady. A headset beats a laptop microphone in a room with echo.
Correct as you go by typing: the text box is a normal editor. When you finish, “Tidy the text” removes filler words and repairs spacing and capitals in one pass, and Undo brings the original back.

Questions and answers

Where does my voice go when I use quick dictation?

Quick dictation uses the speech recognition built into your browser. In Chrome and Edge the browser sends the microphone audio to its vendor’s servers to be recognised; in Safari recognition is handled by Apple. This site does not receive the audio.

How do I dictate without sending audio anywhere?

Use “Record on this device”. It records the microphone in the page and transcribes the recording with the local Whisper model after you press stop. The text arrives when the recording ends, not word by word.

Why does quick dictation not work in Firefox?

Firefox does not ship the speech recognition interface. The page detects that, says so, and offers local recording, which works there.

What does “Tidy the text” change?

It can remove filler words for the language you are dictating in, fix doubled spaces and punctuation, and capitalise the first letter of each sentence. Each option is a checkbox, and Undo restores the text as it was.

Is the dictated text saved?

It is kept in this browser’s local storage so a reload does not lose it. “Clear” removes it.