What does speech to text dictation do?
It listens through your microphone and turns what you say into editable text, using the speech recognition engine inside the browser. Words appear as they are recognised, and the transcript is a normal text box — you can correct a misheard word, delete a sentence or carry on typing by hand without leaving the page.
The language picker matters more than it looks. Recognition is trained per locale, so dictating English into a Spanish setting produces nonsense, and a language with a strong accent needs the right variant selected as well. The list here covers the common cases, including Urdu, Hindi and Bengali, and any locale the browser recognises by name can be used by editing the value.
There is one honest caveat that belongs at the top rather than in a footnote. In Chrome and Edge the audio is recognised on the vendor's servers rather than on the device, which is why it is accurate and also why it needs a network connection. This page cannot change that — there is no in-browser engine that can do the job — so it states it instead, and the recognition happens only while the microphone indicator is on.
How to dictate into the browser
- Choose the language you will speak. Recognition is trained per locale, so this is not a cosmetic setting.
- Press Start listening and allow the microphone when the browser asks. Nothing is captured until you do.
- Speak at a normal pace. Recognised words appear in the box, with the not-yet-final part shown as interim text that may still change.
- Correct anything the engine misheard by typing over it. Editing does not stop the recognition unless you press Stop.
- Press Keep the interim text if you want the last unconfirmed words pinned down before they are replaced.
- Press Stop listening when you are finished, then copy the transcript, download it as transcript.txt, or clear it and start again.
Speech to Text limits and rules
| Item | Limit |
|---|---|
| Browser support | Chrome, Edge and Safari implement the API; Firefox does not, and the tool says so rather than failing quietly |
| Processing | Chrome and Edge recognise speech on their vendor's servers, so a connection is required |
| Languages | Fifteen common locales in the picker, including Urdu, Hindi and Bengali |
| Continuous mode | Restarts itself after the browser stops on a pause, until you press Stop |
| Interim results | Shown as they arrive and replaced by the final text once the engine confirms them |
| Confidence | Reported for the last final result, where the engine provides it |
| Punctuation | Inserted inconsistently — some engines add it, some do not, so expect to edit |
| Audio file | None is produced; only the transcript is kept, and it lives in the page |
When to use it
- Drafting a paragraph or an email faster than you would type it.
- Dictating notes while your hands are occupied with something else.
- Transcribing a spoken passage quickly and then editing it into shape.
- Writing in a language whose keyboard layout you do not have installed.
- Taking minutes or a dictation that has to be searchable text rather than a recording.
Is it private?
Honest version: in Chrome and Edge the audio you speak is sent to the browser vendor's speech service to be recognised, because no in-browser engine can do the job. Nothing is stored by this page — the transcript lives in the text box and the file you download is written from memory — but the recognition itself is not local, and you should dictate accordingly.
Frequently asked questions
In Chrome and Edge it goes to the vendor's speech service over the network. That is a property of the browser, not of this page: the Web Speech recognition API is defined as an interface to a recognition service, and Chrome's implementation calls a cloud endpoint. Safari uses Apple's service with similar implications. Firefox does not implement the API at all. So dictation is fast and accurate and, unavoidably, not private in the way the rest of this site is. Nothing is retained here, but you should not dictate anything confidential.
Because continuous mode is still bounded by silence. The engine ends the session after a pause it considers long enough, which is deliberate — it keeps a forgotten tab from listening indefinitely. This page restarts it automatically while you have not pressed Stop, so in practice dictation continues, but the transcript can be split across sessions and the indicator flickers. When you are finished, press Stop and the microphone is released.
Because recognition and formatting are separate problems. The engine returns the words it heard; whether it inserts a full stop when it hears a falling tone, or a comma at a slight pause, depends on the engine and the language, and several engines do none of it. The old trick of saying period or comma aloud still works in some engines. The reliable answer is to treat the transcript as a first draft and punctuate it yourself, which is faster than fighting the engine.
Every locale the browser recognises, which in practice is a list of a hundred or more language codes. The picker here shows fifteen common ones, including Urdu, Hindi and Bengali, and each is a locale rather than a language — en-GB and en-US are separate recognisers with different training. If your language is not listed, the value can be edited to its BCP 47 code, such as cy-GB for Welsh, and the browser will use it if it can.
This page does not: no audio is stored, and nothing is written to disk. The recogniser receives the audio stream while the microphone is open, and what the browser vendor does with it is governed by their policies rather than by anything here. What you can verify locally is that the page itself keeps only the transcript, and that closing or reloading the tab discards it along with everything else. The microphone permission can be revoked at any time in the browser's site settings.
Not here. The API listens to a live microphone feed, not to a file, so an existing recording has to be played aloud near the microphone — which works, but badly, and it is not a serious method. Transcribing an existing file properly needs a speech-to-text model running somewhere that can read the bytes, either on your own machine or as a service. That is a different kind of tool, and it is not one a browser page can offer without uploading the file.
Related tools
Written and maintained by Md. Nur Hossain Repon.