How to transcribe a voice memo without uploading it anywhere

The web transcription services that rank first for this search all share a shape: you upload a file, it sits on someone’s storage, and the terms of service are the only thing standing between that audio and whatever the operator wants to do with it. For a lot of recordings that is a perfectly reasonable trade. For a lot of others it is not, and people only think about it after the upload finishes.

The short version

• Use an app that calls iOS’s built-in speech recognition rather than its own web API.
• Make sure the language you are transcribing has its on-device model installed.
• Put the phone in Airplane Mode before you transcribe.
• If text comes back, it was produced locally. There was nothing to upload it to.

Step 1: get the audio onto the phone

If you recorded it on the same iPhone, skip this. If it came from elsewhere — a handheld recorder, a laptop, a colleague — move it in a way that does not itself involve an upload. AirDrop is the cleanest for Mac and iPhone. A USB cable and the Files app works for anything. Emailing it to yourself technically works but puts a copy on two mail servers, which rather defeats the point.

Apps that support importing audio can then pick the file up from the Files app. Voice Studio does this; so do several others.

Step 2: make sure the local model is there

On-device speech recognition needs a language model that has actually been downloaded to the phone. The reliable way to trigger that: Settings → General → Keyboard, add the language, and make sure Dictation is enabled. Leave the phone on Wi-Fi and charging for a while. iOS fetches these assets on its own schedule and gives you no progress bar.

Then test it: open Notes, tap the microphone on the keyboard, and dictate a sentence in Airplane Mode. If the words appear, the local model is present. If nothing happens, it is not, and no app can conjure it.

Step 3: transcribe with the network off

This is the step that turns a promise into a fact. Enable Airplane Mode and confirm Wi-Fi is off too — iOS remembers Wi-Fi separately and will happily reconnect. Then run the transcription.

Working offline does two things at once. It guarantees no audio was transmitted, because there was no route out. And it flushes out apps that quietly fall back to a server: those will fail here rather than succeed, which is exactly the information you wanted.

Step 4: check the result before you trust it

Local models are good, not perfect. Names, jargon, numbers and overlapping speech are where they slip. Read the transcript against the audio once, at speed, and fix what matters. A transcript you have not checked is a rumour with timestamps.

Editing matters more than raw accuracy for most uses. If the app lets you correct the text and keeps your correction, one pass is usually enough to make the transcript searchable and quotable.

What you cannot do this way

Speaker labels ("Alice:", "Bob:") are not something the on-device path provides. Neither is summarisation — anything advertising an AI summary is calling a model somewhere, and that means the text is going out even if the audio did not. If a tool offers you a summary while claiming to be fully offline, that claim deserves a closer look.

Doing this in Voice Studio

Record in the app or import an existing audio file, then transcribe. The on-device path is attempted first every time. With the network off, that is the only path available, so a transcript you get in Airplane Mode was produced on your phone. There is no account to create and no server of ours involved at any point; the recordings and the text sit in the app’s own storage, and you can export either one whenever you like.

Common questions

Can the built-in Voice Memos app transcribe offline?

Recent iOS versions added transcription to Voice Memos and it uses the same system speech recognition. Whether a given transcription runs locally still depends on the language and whether its on-device model is installed.

Does transcribing offline drain the battery faster?

Running a speech model on the phone does use the CPU and the neural engine, so a long file will warm the phone up. Plugging in for a long transcription is a reasonable habit.

How long a recording can this handle?

There is no fixed ceiling from the framework; implementations split long audio into chunks. Practically, the limits you hit first are free storage and how long you are willing to wait.

Is a transcript stored on the phone encrypted?

App storage on an iPhone is covered by the device’s own file protection, which is tied to your passcode. That is what protects it — not anything an individual app adds on top.