🎙️ Speech to text, speakers included

transcribr-web

Transcription with speaker detection, entirely in your browser. Nothing you pick leaves your computer. Desktop Chrome or Edge.

🔒 100% local: your audio is never uploaded. Only the transcription engine itself (not your files) is downloaded, once, from Hugging Face.

Speakers

0 = detect automatically

Language

Blank = detect automatically

Translate to English

Drop audio or video files here
or click to choose files

Options above apply to files as they start processing, so you can still adjust them per file.

Settings

Model

Which model should I pick?

	Fast	Standard	Accurate
Download, once	32 MB	190 MB	514 MB (1.6 GB on a graphics card)
Speed on a laptop	Several times faster than the recording	About as long as the recording	Two or more times longer than the recording on a processor; much faster on a graphics card
Mistakes	Misses names, jargon and accents; drifts on noisy audio	Good on clear speech in common languages	Fewest. Holds up with accents, noise, crosstalk and less common languages
Pick it for	A quick rough draft, or an old or slow computer	Most interviews and meetings	Anything you'll quote or publish, or hard audio

All three are the same Whisper model at different sizes, so they share languages and features; bigger just means fewer mistakes and more waiting. Not sure? Start with Standard, and rerun a file on Accurate if names or words come out wrong. Speaker detection is the same whichever you pick.

On a graphics card, Accurate uses Whisper large-v3-turbo, a newer and larger model that is usually more accurate still. It can't translate, so with Translate to English on, Accurate runs on the processor.

Run on

Appearance

Saved in this browser.