Transcription of recordings

You send us a recording and get back what was said, written down, in the language it was spoken. Text, subtitles, or the raw timings for your own tools.

Get startedSee pricing
  • Fifteen languages
  • You pay for speech, not silence
  • No monthly fee

Who this is for

Anyone with a recording and no wish to type it out. A council session that has to become minutes. A sermon that goes on the parish website. Interviews somebody has to quote from accurately. Lectures that students want to search rather than rewatch.

How it works

  1. 1

    You upload the file. Any usual format — wav, mp3, m4a, ogg — and video too, we take the sound out ourselves.

  2. 2

    We find where the speech is before doing anything else, and tell you how many minutes of it there are and what that will cost. You know the price before it queues.

  3. 3

    It runs when no live room needs the card, so a recording never delays somebody's event.

  4. 4

    You get an email when it is done, and the result waits in your account.

What you get

  • The full text, in the language it was spoken.
  • Subtitles in SRT and VTT, with their timings.
  • The raw turns as JSON, with start and end, for feeding into something else.
  • Only the speech: silences, applause and dead air are not in there and are not billed.
  • The file you sent is deleted the moment the job is finished.
  • Results kept in your account for fifteen days.

Where the limit is: four hours of audio, one hour of video, and 800 MB per recording, three jobs queued at a time. A longer session goes up in several pieces. A recording takes around a tenth of its own length to come back — an hour of audio is about six minutes of waiting, when the machine is free.

What it costs

€2 per hour of audio

Speech, not duration. A two-hour meeting where people actually talked for fifty minutes costs fifty minutes of speech, not two hours.

See pricing

Questions

What accuracy should I expect?
It depends almost entirely on the recording. A lapel mic in a quiet room transcribes close to perfectly; a phone on a table in a hall with an echo will lose words, and no engine fixes that afterwards.
Does it punctuate?
Yes — full stops, commas and capitals. It is written to be read, not a wall of lowercase.
Can I send video?
Yes. mp4, mov, mkv, avi and webm. We take the audio out and work on that; the video is not kept.
Is my recording used to train anything?
No. It is deleted when the job finishes, and nothing of it is kept beyond the transcript in your own account.
What if the audio is in more than one language?
The engine detects what is being spoken. If people switch language mid-recording, tell us — that case is worth testing on your own audio before you rely on it.

The rest of what happens to a file

The same file, the same engine and the same pot of credits. Almost nobody needs just one of these: whoever transcribes a session ends up wanting to know who spoke.

Meeting notesWhat was decided, who takes each task, and the minute it was saidSubtitlesSRT and VTT with their timings, ready to drop on a video.Translate recordingsThat same recording, in every language you need.Translate documentsPDF, Word and six more formats, translated whole.Who said whatWho said what, marked in the transcript.

The words only used in your house

An organisation's own names — people, streets, references, job titles — are exactly what a general model has never heard, and where it fails: it swaps them for an ordinary word that sounds similar. Write them once and they stop failing. You write them in the job and they hold for that recording.

  • It is a list, not training: you write it and it works from the first sentence.
  • It collects nothing. Not your audio, not your text, not one second of anything.
  • And it is included. Not an extra plan, not a line on the bill.

Try it with your worst recording

Not the good one. The one with the echo and the coughing, because that is the one that tells you whether this is any use to you.

Get started