About

You upload a recording or paste a link, and you get back the words with timestamps — down to the individual word — as plain text, SRT or VTT. It handles interviews, podcasts, lectures and recorded meetings, in 99 languages, detected on their own.

Three transcripts are free and need no card. After that a plan costs 49 or 89 kr a month.

Ari Batak, in Sweden. One person, not a company with a support department. Write to skribiasupport@gmail.com and the reply comes from the same person who wrote the code.

Transcription runs on a graphics card we own and run ourselves, in Sweden. Your audio is not forwarded to OpenAI, Google, Amazon or any other transcription service, and it is not used to train anything. That is the whole reason the price is what it is: there is no per-minute fee to pass on to you.

It also means the queue is real. One job runs at a time, at roughly eight times faster than the recording is long — about three minutes of waiting for a half-hour interview, when nobody is ahead of you.

Music. The engine listens for speech, and sung vocals over a mix leave it almost nothing to hold on to — a song can come back nearly empty. It is also not a translator: it writes down what was said, in the language it was said in.

And it is not a company with a night shift. If the machine goes down at three in the morning, it stays down until morning.

It is short, and every retention period in it is a real setting in the running code rather than a promise made in the abstract. It says where your audio is processed, what is kept, for how long, and what we can never see — including what a leaked database would and would not reveal.

Read it before you upload something that matters: the privacy page, and the terms.