Can you re-transcribe a voice memo after picking the wrong language on iPhone?

A transcript that is garbled top to bottom looks like the recognizer had a bad day. Usually it did not — it was simply told to listen for a language that was not the one being spoken, and every guess it made afterward was built on that wrong premise. That is a different failure from the ordinary kind, and it has a cleaner fix than the ordinary kind does too.

Why the wrong language breaks everything, not just a few words

iOS speech recognition does not detect which language is being spoken — it is told which one to expect, and it spends the entire pass trying to fit what it hears into that language's sounds and vocabulary. Point it at English while someone is speaking Spanish, and it is not making a series of small mistakes the way it does with an unfamiliar name or a fast list of numbers. It is mapping foreign phonemes onto the nearest English-shaped words it can find, for every single word, which is why the result reads as nonsense rather than as a transcript with a few rough patches.

This is worth separating from the more familiar kind of bad transcript, where the language was right and a handful of names or technical terms came out wrong. That version is a targeted fix — play back the one section, correct the one word. A wrong-language transcript is not a handful of errors sitting inside otherwise-correct text; it is wrong at the root, and no amount of editing a word here or there gets you back to something usable.

Regional variants count as a language mismatch too

This is not only about switching between, say, French and Japanese. iOS treats regional variants of the same language as separate options a transcription setting can point at — English (US), English (UK) and English (India) are distinct entries, not accents one model quietly absorbs. A transcript that is merely disappointing rather than complete nonsense, with an unusual number of near-misses on ordinary words, is worth checking against this before assuming the recording was noisy: the language might be broadly right and the regional variant wrong.

What actually needs to change

The transcription language is its own setting, chosen separately from anything else about the app, and it has nothing to do with the language the app's own buttons and menus are displayed in — those are two independent choices that do not move together. Fixing a wrong-language transcript means finding that one setting and pointing it at the language actually spoken in the recording, not changing anything about the phone as a whole.

What happens when you run it again

The recording itself was never touched by any of this. Audio is audio regardless of which language a later transcription pass expects, so once the setting points at the right language, running transcription again on the same file is a completely ordinary attempt — the same on-device pass first, the same fallback to Apple's server only if that pass comes back empty or errors, exactly as it would for a recording transcribed correctly the first time. Nothing about having gotten it wrong once puts the second attempt on a different path.

The transcript itself works as a single field attached to the recording, not a running history of past attempts — editing it by hand, the ordinary way of fixing a wrong word, replaces what was there rather than keeping the old version alongside the new one. A fresh transcription pass behaves the same way: once it finishes, that is what the recording's transcript holds. If there is any reason to keep the original, nonsense version around — comparing the two, or just curiosity — export it as text before running the correct pass, because the field itself does not keep both.

What a language fix does not fix

Getting the language right removes one specific cause of a bad transcript. It does not touch the others. A recording made too far from the speaker, or in a noisy room, still degrades the signal before any recognizer — in any language — ever sees it, and the usual errors around names, jargon and fast speech still apply once the language is correct. If a transcript is still rough after fixing the language, the remaining problems are the ordinary kind, worth a normal read-and-correct pass rather than another setting change.

How Voice Studio handles this

The transcription language is a free setting, chosen from among 38 supported languages, separate from the app's own 19 interface languages — changing one never moves the other. Transcribing a recording always attempts the on-device pass first and only falls back to Apple's server if that pass comes back empty or errors, regardless of whether this is the first attempt on a recording or a second one after correcting the language. The transcript is editable text sitting on the recording, not a version history, so running transcription again puts a fresh result in that same place — export the recording's transcript first if there is a reason to keep what the wrong language produced. None of this touches the audio itself, and none of it is held back for Pro: recording, transcription, markers, search and export are unrestricted in the free app.

Common questions

Does re-transcribing fix a transcript that is only slightly off, or just completely wrong languages?

Both, for the same underlying reason. iOS speech recognition is told which language to expect rather than detecting it, so even a regional mismatch — English (US) set while someone spoke English (UK), for instance — degrades accuracy, not just a full language swap.

Will transcribing a recording again change the audio file?

No. Transcription only reads the audio to produce text; it never modifies the recording. The .m4a stays exactly as it was captured no matter how many times it gets transcribed.

Do I need to delete and re-record to fix a wrong-language transcript?

No. The recording is unaffected by which language a transcription pass expected. Change the transcription language setting and run transcription again on the same file.

Does a second transcription attempt keep the first, wrong version around anywhere?

Not as a separate copy. The transcript is a single field on the recording, so a new pass replaces what was there, the same way manually editing a word replaces it. Export the original text first if you want to keep it.