Transcription runs on your own computer. Drop a file in, watch the text appear, download TXT, SRT or VTT. Nothing is uploaded, nothing is metered.
Nothing was downloaded. The tool works on desktop Chrome or Edge 113 or newer with a graphics card that supports WebGPU. Firefox, Safari and phones are next on the list.
Nothing was uploaded. If this keeps happening, tell us which browser and file type you used.
Most tools send audio to a server and ask you to trust a policy. Here there is no request to make: the speech model runs inside the browser tab, on your graphics card.
Thirty minutes a month, then a wall. Local processing costs us nothing per file, so there is no meter, no queue and no account.
Plain text, SRT and VTT all download immediately, with timestamps, without an email address.
The first run fetches the speech model, about 200 MB, into your browser cache. Every later file starts instantly.
The browser reads the audio a few seconds at a time, so an hour-long meeting recording does not need an hour of memory.
WebGPU runs the model on your graphics card, faster than the recording plays back, and the text appears on screen as it goes.
Search inside the transcript, copy it, or download TXT, SRT or VTT. Close the tab and it is gone from this site, because it was never here.
An SRT file is four lines of plain text per subtitle. Here is the exact format, three ways to make one, the timing rules broadcasters use, and the small mistakes that make YouTube or VLC reject the file.
What an hour of transcription really costs on Otter, Rev, Descript, the cloud APIs behind them, and a model running on your own laptop. Prices checked September 2026.
Three exports, three jobs. A short decision table for transcripts and subtitles.