How to transcribe a voice memo in a different language on iPhone

People run into this the same way every time: they record something in, say, Portuguese, on a phone whose menus are in English, and the transcript comes back empty or nonsensical. The instinct is to blame the app. Usually the real cause is a language setting that never got changed from whatever it defaulted to.

Two different language settings, easy to conflate

An app like this has two separate notions of language, and they do not move together. One is the interface language — the language the buttons, menus and screen text are written in. The other is the transcription language — which language the speech recognizer is told to listen for in a given recording. Voice Studio supports 19 interface languages and, separately, 38 languages for transcription. Changing one does nothing to the other.

That gap matters in exactly the situation described above: a phone set to English can still transcribe Portuguese, Japanese or Arabic, because the interface language was never the thing controlling that. What controls it is the transcription language setting, and if that has been left on the wrong language, the recognizer is listening for the wrong sounds no matter how correctly you set anything else up.

How iOS actually picks a language to listen for

It does not detect the language — it is told

A speech recognition request on iOS specifies a language up front. The system does not sample a clip, work out what language is being spoken, and then transcribe it — it starts from the assumption that a particular language is coming and matches sound against that language's model the entire time. This is a property of how the underlying framework works, not a limitation specific to any one app.

What happens with the wrong language selected

The recognizer still tries. It matches the sounds it hears against the phonemes and vocabulary of the language it was told to expect, which for a genuinely different language produces text that is either empty, or made of real words in the wrong language that do not correspond to anything actually said. It rarely fails loudly — a transcript that looks superficially like normal sentences but means nothing is a classic sign the language setting was wrong, not that the audio was bad.

When your language or dialect is not covered

Not every language variant that exists has its own entry. Broadly spoken languages tend to have several regional options; smaller languages may have one, or may not be covered at all. If a specific dialect is not listed, the closest broader entry for that language is usually the better choice than leaving the setting on something unrelated — a near match against the wrong-but-similar language will still outperform a match against a completely different one.

Recording something that switches languages partway through

A recognizer is matching against one language for the whole request, so a conversation that moves between two languages — common in bilingual households, some business meetings, language-learning practice — will transcribe cleanly in whichever language matches the setting and poorly in the other. There is no partial fix from the audio side; the practical approach is to pick whichever language dominates the recording and expect the other stretches to need a manual pass afterwards.

On-device or the server, and how the language affects that

Separately from which language is set, there is the question of whether recognition runs locally on the phone or on Apple's servers — Voice Studio always attempts on-device first and only falls back to the server if that local pass comes back empty or errors. Language coverage for the on-device path is narrower than for transcription overall, and it is not something the app controls; some languages that transcribe reliably do so because the fallback quietly reached the server, not because everything ran locally. That question is covered in more detail in the companion guide on on-device versus cloud transcription — worth reading if privacy for a specific language matters to you, since the answer genuinely varies by language.

How Voice Studio handles this

The transcription language is a setting you choose, separate from the app's 19 interface languages, from among the 38 languages the app supports for transcription. It is not one of the things held back for Pro — like recording, markers and search, it is there in the free app. If a transcript keeps coming back empty or nonsensical for a recording you know is clear, that setting is the first thing worth checking, well before assuming the recording itself is unusable.

Common questions

Does changing the app's interface language also change the transcription language?

No. They are independent settings — 19 interface languages and 38 transcription languages in Voice Studio — so changing one does not move the other.

Can iPhone automatically detect which language is being spoken?

No. Speech recognition on iOS is told which language to expect before it starts and matches sound against that language the whole way through. It does not sample the audio first and guess the language.

Why does my transcript come back as real words that make no sense?

That is the classic sign of the transcription language being set wrong rather than the audio being bad — the recognizer matched what it heard against the wrong language's vocabulary and produced text anyway.

What happens if I record a conversation that mixes two languages?

Whichever language the setting matches will transcribe reasonably well; the other language will not, because the recognizer is only listening for one language per recording. Expect to correct those stretches by hand.