Automatic language detection

How Sprecho works out which language you just spoke, why detection runs on every dictation, and what multilingual users should do to get the cleanest results.

Updated August 23, 20263 min read
FreeProEnterprise

Sprecho detects the language of each dictation on its own. There is no "language mode" to enter and no button to press before you switch from German to English — you speak, and the text comes back in the language you used.

For anyone who works in more than one language, this is the whole point: the tool adapts to the sentence instead of the sentence adapting to the tool.

What detection actually does

Two things depend on the detected language:

  1. Transcription — the words themselves.
  2. Formatting — punctuation, capitalisation and phrasing conventions differ between languages. German capitalises nouns; French puts a space before a colon; English does neither.

Both follow the language of the clip you just recorded, which is why a German paragraph and an English paragraph in the same document each come out correct.

When to prefer one language per dictation

Detection is decided per recording, not per sentence. That has one practical consequence:

If you speak two languages inside a single recording, one of them wins for the whole clip. Release the key between languages and you get two clean results instead of one compromised one.

So, if you are writing a bilingual email, dictate the German part, let go, click below it, and dictate the English part. It takes half a second longer and removes the only real failure mode.

What helps detection

  • Speak for at least three seconds. Detection is reliable from about that length. "Yes" or "OK" gives it almost nothing to judge.
  • Start with a full sentence rather than a name. A recording that begins with a proper noun in another language is the hardest case there is.
  • Avoid heavy background speech. Another conversation in a different language in the room is exactly the kind of input that confuses a detector.

Where the setting lives

Settings → Dictation Languages has an Auto-detect language toggle, on by default, and a list of the languages you speak. The list is worth keeping accurate — it tells the app who you are and shows up in the picker when you import an audio file, where you can also pick a language explicitly.

For everyday dictation, leave it alone. There is nothing to maintain.

When you would rather be explicit

There is one place where naming a language is genuinely useful: transcribing an existing recording. The Import audio screen has a language dropdown, set to Detect automatically, which you can change if you already know what is on the tape. See Transcribe an existing audio file.

For multilingual users

A few habits that make a bigger difference than any setting:

  • Keep the personal dictionary in whichever language the term belongs to. A client called "Nordbau Ingenieure" should be added once and it will be spelled correctly in both your German and your English dictations.
  • Don't slow down when you switch. Over-enunciating a foreign name usually makes it harder to recognise, not easier.
  • Watch the first sentence after a switch. If a language switch is going to go wrong, it goes wrong at the boundary, and a quick glance catches it.

Frequently asked questions

Every recording is analysed for its language before the words are transcribed, and the result is used for both the transcription and the formatting. Nothing is set in advance, so switching languages needs no action from you.

Was this article helpful?

Still stuck?

Tell us what you are trying to do and we will walk you through it.

Related articles