Why does my iPhone transcript write numbers as digits instead of words?

Say "I paid twenty five dollars for it" into a transcribing app and the transcript is likely to come back as "I paid $25 for it" or "I paid 25 dollars for it," not a word-for-word reproduction of what left your mouth. The instinct is to read that as an error, as if the recognizer heard "twenty five" and somehow wrote "25" by mistake. It did not mishear anything. It heard the number correctly and then made a deliberate choice about how to represent it in writing — a separate step from recognition itself.

Recognition and formatting are two different jobs

Turning speech into text is not one operation — it is at least two, done in sequence. The first is recognition: matching sound against a model to work out which words were actually said. The second is formatting: deciding how those words should look on the page. For most words the two stages agree completely, so the distinction is invisible — "the" gets recognized as "the" and written as "the," nothing to notice. Numbers are one of the few categories where the two stages diverge on purpose, because the words a person says out loud and the way those same words are normally written down are not actually the same thing.

Why digits win

Written English almost never spells a number out the way it is spoken, outside a handful of specific contexts — a check, a formal document, a lone digit in the middle of a sentence. A price, a date, a phone number, an age, a percentage: in ordinary written text these are digits, essentially every time. A speech recognizer's formatting stage is trained on that same body of ordinary written text, so it reproduces the convention that text actually uses rather than a literal spelling-out of the syllables you spoke. This is not unique to any one app — it is the same behavior iOS dictation uses in Messages or Notes, turning "meet me at seven thirty" into "meet me at 7:30" without anyone asking it to.

Where else it shows up

• Dates and times — "the third of May" or "half past two" tend to come back as a written date or a clock time rather than the words that were said.
• Money and quantities — an amount spoken as words usually lands as a number, sometimes with a currency symbol attached.
• Phone numbers and addresses — a string of spoken digits comes back as digits, grouped the way a written number normally is, not as separate number-words.
• Small counts inside a sentence — this is the one that varies most. A lone "three" in casual speech can go either way depending on context, unlike a price or a date, which formats consistently.

The one place this habit is worth checking, not trusting

Formatting a correctly heard number as a digit is harmless — it changes how the number looks, not what it means. The real risk sits one step earlier, in recognition itself: numbers are one of the categories a speech recognizer gets wrong most often, right alongside names and jargon, because a mumbled or quickly spoken digit has fewer contextual clues to lean on than an ordinary word. A digit that looks clean and confident on the page is not proof it is the right digit — it only means whatever number the recognizer landed on got formatted well. For anything a wrong number would actually matter for — a dosage, an address, a total, a phone number — check it against the audio specifically, the same way a name gets checked, rather than trusting it because it looks tidy.

Is there a setting to turn this off?

Not really, and it would not be an app-level setting even if there were one. The formatting happens inside Apple's speech recognition itself, as part of turning recognized words into the text it hands back — it is not a separate pass a transcribing app runs afterward and could offer a toggle for. If a document specifically needs numbers spelled out as words rather than digits, the transcript is ordinary editable text once it exists, so that is a correction made by hand for the handful of numbers where it actually matters, rather than something a setting is likely to solve.

How Voice Studio handles this

The transcript Voice Studio shows is exactly what Apple's speech recognizer returns, formatting included — there is no separate pass that rewrites numbers, dates or anything else afterward. Transcription attempts the on-device recognizer first for whichever language is selected, and only retries against Apple's server if that local pass comes back empty or errors, so the same digits-versus-words behavior applies either way. Because the transcript is plain, editable text exported as TXT or JSON, fixing a genuinely wrong number — as opposed to a correctly heard one that simply got written as a digit — is a normal edit, the same as correcting a misheard name.

Common questions

Is it a mistake when my transcript shows "25" instead of "twenty five"?

No, that is normal formatting, not an error. Written numbers are shown as digits in ordinary text almost every time, and speech recognition follows the same convention rather than spelling out the words you spoke.

How do I tell a formatting choice apart from an actual misrecognition?

Check the number against the audio. Formatting only changes how a correctly heard number looks — 25 versus twenty five — it does not turn "seven" into "seventeen." If a number looks flatly wrong or does not match what was said at all, that is a separate mishearing worth fixing, not just a style choice.

Can I make my transcript spell numbers out as words instead of digits?

There is no built-in toggle for this — the formatting comes from the speech recognizer itself, not a setting in the app. Edit the transcript afterward for the specific numbers where the spelled-out form matters.

Does this happen with dates and prices too, not just counting?

Yes, generally. It covers most kinds of numbers — dates, times, prices, phone numbers — since those normally appear as digits in written text anyway, not just numbers said while counting something out loud.