Dictation online without upload

No. The microphone is only switched on once you allow it in the browser, and the recording is recognised with Whisper on your device only – neither sound nor text leaves it. The only downloads are the speech model (from Hugging Face) and the runtime (from jsDelivr), once; the browser keeps both for later visits.

Whisper understands about 100 languages. The common ones can be selected; otherwise “Detect automatically” works out the language for every section itself. A fixed language is more reliable, especially for short sentences. You can also change the language during a dictation.

Whisper usually adds full stops, commas and question marks itself – speak naturally and let your voice drop at the end of a sentence. You can also say the punctuation: “period” (or “full stop”), “comma”, “question mark”, “exclamation mark”, “colon” – after a full stop, question or exclamation mark the next word starts with a capital. “New paragraph” starts a paragraph, “delete last sentence” takes the last sentence back. The commands themselves do not appear in the text. Note: the word “period” is always taken as punctuation, even in “a period of time”. Afterwards you can click any line and correct it by hand.

“Standard” (Whisper base) is the best choice for reading along: with a graphics card you see a preview while you speak. “Fast” (Whisper tiny) responds quickly even without a graphics card but makes more mistakes. “Accurate” (Whisper large-v3-turbo) needs several seconds per sentence – so “Standard” types along during the dictation, and “Transcribe again” redoes the recording with “Accurate” afterwards. Names, technical terms and dialect remain error-prone: please proofread the text.

Whisper recognises whole phrases better than single words. So the tool waits for a pause of a good half second and then sends the sentence to the model. With a graphics card (WebGPU in Chrome and Edge) the finished sentence is there one to two seconds later, and a preview runs along before that. Without a graphics card a single processor core does the work; on a phone the text may then lag a few seconds behind.

You need a current browser with microphone access. Dictation runs most smoothly in Chrome and Edge on a computer, because the graphics card helps there. A headset or a microphone close to your mouth improves the result a lot; background noise and several voices at once get in the way.

A dictation can last up to 60 minutes on a computer and 30 on a phone. You can stop it and continue later – the new part is appended. The recording only lives in the memory of the page: you can save it as WAV; when you close or reload the page it is gone, and so is the text. So copy or save it first.

Yes. Switch from “Microphone” to “File” at the top and choose an audio or video file – or use the “Transcribe audio” tool, which adds timestamps, speaker labels and subtitle export (SRT/VTT).