Skip to content

Quick answer

How do I translate audio or a recording into English text?

Transcribe the recording, then translate the transcript into English (or Arabic) with a free account. There is also a one-step option for English only: set the task to "Translate to English" when you start a transcription. Both are automatic and both make mistakes, so treat the result as a draft and check names and numbers against the original. Nothing is sent to an outside translation service: the translation runs on our own server.

Updated

The two routes inside TryScriptorium

  1. What it does
    Translate while transcribing: The speech model listens to the recording and writes English text directly
    Translate a finished transcript: A separate, small translation model translates the text of a transcript you already have
  2. Languages out
    Translate while transcribing: English only
    Translate a finished transcript: English or Arabic
  3. Languages in
    Translate while transcribing: The spoken languages the site supports (33 plus automatic detection)
    Translate a finished transcript: Any language the transcript is in that the translation model knows (about 100)
  4. Where you start it
    Translate while transcribing: Account: Advanced options, Task, "Translate to English", before you upload
    Translate a finished transcript: Account: the Translate view of a finished transcript
  5. What you keep
    Translate while transcribing: Only the English text
    Translate a finished transcript: The original transcript and the translation, with the same timestamps, side by side or one at a time
  6. Export
    Translate while transcribing: TXT, timestamped TXT, SRT, VTT, Markdown, JSON and CSV
    Translate a finished transcript: TXT, SRT, VTT and JSON of the translation
  7. Allowance
    Translate while transcribing: Counts as a normal transcription against your monthly minutes
    Translate a finished transcript: A daily allowance for translations that depends on your plan; see the pricing page
  8. Good for
    Translate while transcribing: A quick English version of one clear recording
    Translate a finished transcript: Arabic as a target, a translation you can compare line by line, or a transcript you have already corrected

Translating a transcript you have already corrected usually gives a cleaner result than translating while transcribing, because a wrong word in the original is fixed once, before it is translated. That is the reason to prefer the second route for anything that matters.

Steps (the transcript route)

  1. Transcribe the recording. Drop the file on audio to text. One file up to 15 minutes and 25 MB, 3 tries a day, no account. Choose the spoken language if you know it; automatic detection works for most clear recordings.
  2. Keep the transcript. Create a free account from the transcript screen. The transcript you just made comes with you; translating needs an account.
  3. Open the Translate view. In the transcript, choose Translate, then English or Arabic. The original stays; the translation is kept beside it with the same timestamps.
  4. Read both and fix names. Read the translation next to the original. Names, numbers and technical words are where machine translation goes wrong first.
  5. Export. Copy the translation, or download it as TXT, SRT, VTT or JSON.

How good is it?

We make no quality claim and publish no translation accuracy figure. The translation model is small, on purpose: it runs on our own machine so that your text does not go to a translation service. That makes it rougher than the large commercial translators, especially on idiom, names, short fragments and mixed-language speech. Every transcript is an automatic draft already, and translating adds a second place for mistakes, so errors can compound: a misheard word is translated as the wrong word. For anything you will publish, sign or act on, have a person who knows both languages read it.

What it does not do

  • Translate into any language other than English or Arabic.
  • Translate live while someone is speaking, or dub the audio into another language.
  • Offer a human translator. There is none.
  • Translate a very long transcript in one go: a transcript over 60,000 characters is refused, so split a very long recording first.
  • Translate on the no-account preview. The preview transcribes; translating needs a free account.

Privacy

The translation model is installed on our server and runs there; the text is not forwarded to a translation API. The recording and the transcript follow the same rules as everything else: guest files are deleted after 24 hours, account files stay until you delete them. Details on the privacy page.

Related

Questions

Can it translate audio to English in one step?
Yes, for English only: with an account, set the task to "Translate to English" before you start. The speech model writes English text directly from the speech.
Which languages can a transcript be translated into?
English and Arabic. A transcript in any language the translation model knows (about 100) can be translated into either.
Does it keep the timestamps?
Yes. The translation uses the original's timestamps, so SRT and VTT exports of the translation line up with the audio.
Is the translation sent to an outside translation service?
No. It is made by a small model running on our own server, and the text is not forwarded to any translation service.
Is translation free?
You can try a transcript with no account (one file up to 15 minutes and 25 MB, 3 tries a day, no account). Translating needs a free account, which has a daily allowance of translations; paid plans have a bigger one. See the pricing page.
Is the transcript always right?
No. It is an automatic draft, and no accuracy percentage is promised. Noise, overlapping voices, accents and technical terms cause mistakes, so read it against the audio before you rely on names, numbers or quotes.

Try it with your own file

Upload a recording and read the transcript.

More pages like this