SCRIBEGRABOpen the tool →

About ScribeGrab

A small, independent transcription tool that runs on hardware we actually own — which is the whole reason it can be free.

What this is

ScribeGrab is a free audio-and-video transcription tool. You upload a recording, our server transcribes it with Whisper large-v3 — one of the most accurate open speech-recognition models available — and you download the result as a plain-text transcript plus time-coded SRT and VTT subtitle files. There's no account, no watermark, no daily cap, and no per-minute meter running in the background.

It's not run by a startup with a sales funnel. It's a small independent project, built and maintained by the same tiny team behind StemGrab (stem separation) and PicReviver (photo restoration).

Why it's free — the honest version

Most transcription services pay a cloud provider for every minute of audio they process, so they have to charge you per minute or cap you at a few files a day. We don't have that bill: ScribeGrab runs on our own GPU, in our own machine. Once the hardware is paid for, the marginal cost of transcribing your file is mostly just electricity.

That electricity is paid for by the ads you see around the tool. That's the entire business model: ads keep the GPU humming, and in exchange the tool stays free for everyone, with no daily cap and no watermark. No premium tier, no "upgrade to unlock SRT export" tricks.

Privacy: your file doesn't stick around

Transcription only works if you trust us with your recordings, so the rules are strict and simple:

• Your uploaded file is deleted the moment processing finishes — success or failure.
• The generated transcript and subtitle files are wiped within 45 minutes, so download them right away.
• Nothing is stored long-term, shared with anyone, or used to train models.
• Everything runs on our own server — your audio is never forwarded to a third-party transcription API.

The full details are in the Privacy Policy.

What it's good at (and what it isn't)

Clear speech — interviews, lectures, podcasts, voice memos, screen recordings — transcribes very accurately in around 100 languages, auto-detected. Heavy background noise, people talking over each other, and very thick accents are harder for any speech-recognition system, ours included; we've written an honest guide about how accurate AI transcription really is. Music won't transcribe — for splitting songs into stems, use StemGrab instead.

Questions or problems?

Found a bug, got a weird result, or just want to say something? Use the contact form — messages go straight to the people who built this.

Try the tool →