Browse all documentation
Customising

Languages and mixing them

How VoiceSnap Pro detects the language you speak, when to pick one yourself, and what happens when you mix languages.

Detected automatically, or chosen

VoiceSnap Pro transcribes 50+ languages. Settings → Language and microphone has one Language setting: leave it on automatic detection and the language is worked out from each dictation, or pick yours when detection gets it wrong.

A short dictation — a two-word search query — carries little to go on. If you dictate mostly in one language and detection now and then guesses another, picking that language is the fix.

Mixing languages

With detection on, you can dictate one message in English and the next in German without touching anything.

Switching inside a single dictation is harder. The model may follow you, keep the whole dictation in one language, or carry a borrowed word across correctly, and the result varies from one language pair to another. Read mixed-language text before you send it.

  • Names and product names are best left to personal vocabulary, which applies whatever language you are speaking.
  • If one language keeps winning when it should not, pick the language for that stretch of work and switch back to detection afterwards.

Punctuation follows the language

Punctuation and capitalisation come from the same model that recognises the words, so they follow the conventions of the language you spoke rather than English rules applied to other alphabets.

Right-to-left text such as Arabic and Hebrew is pasted in logical order, so it displays correctly in any field that supports it.

A voice note in the wrong language

Re-transcribe…, in a note's ⋯ menu, runs its recordings through transcription again in a language you pick, or with detection. The text it replaces stays in version history, so trying again costs nothing.

What VoiceSnap Pro will not do

Dictation transcribes; it does not translate. Speak French and you get French text, so a mis-detected word shows up as an obvious typo rather than a silent mistranslation. When you do want a translation, run the Translate to… AI action on a note afterwards.

Accuracy also varies by language and accent — the widely spoken languages have far more training data behind them than the rarer ones. If a language matters to your work, test it with the kind of material you actually dictate before you rely on it.

Still stuck?

If this page did not solve it, send us the app you were dictating into and which of the three checks failed — no mic pill, no level meter, or the wrong app named. That is usually enough to answer in one reply.

Back to documentation