Voice notes that never reach a server

For work where sending a recording to a third party is not a preference to weigh, but a line you cannot cross.

The question that stops most dictation apps

Lawyers, doctors, psychologists, notaries, social workers, HR staff, journalists protecting sources — all of them dictate notes, and all of them run into the same wall. Dictating a note about a client into an app that uploads audio to a server means a third party now holds that recording.

Depending on your profession that can mean a confidentiality breach, a data protection problem, or simply a risk nobody wants to carry. And it is rarely worth the argument with a compliance officer.

So most people fall back on typing it later, which costs an hour a week, or on a voice memo they never revisit.

Why "we don't look at your data" is not the same answer

Most privacy policies promise good behaviour: we encrypt, we do not train on your data, we delete after processing. These may all be true. They are still promises about what someone chooses to do with data they hold.

There is a structurally different answer: the data never arrives. If the transcription happens on your phone, no third party holds the recording, because none ever received it. Nothing to breach, nothing to subpoena, nothing to delete.

The strongest privacy guarantee is not a policy. It is the absence of a server.

What to check before trusting an app

1. The airplane mode test. Enable airplane mode, turn off Wi-Fi, dictate. If it works, processing is local. If it fails, it was not.
2. Is an account required? An app that makes you sign up has a server, and a record of you.
3. Does the AI feature also run locally? This catches people out. An app can transcribe on-device and then send the text to a cloud AI for the summary. The transcript is the sensitive part — check both halves.
4. Where do the notes live? Cloud sync is convenient and means a copy exists elsewhere.
5. What does the App Store privacy label say? "Data Not Collected" is a statement Apple holds the developer to.

The trade-offs, stated honestly

On-device processing costs you something, and you should know what before choosing it:

• A model that fits on a phone is smaller than one running in a datacentre. On clean dictation the difference is small; on difficult audio it is real.
• No cloud sync means notes live on one device. Back up accordingly.
• First launch needs a network connection to download the model files. Once.

For dictating notes between appointments, these are minor. For transcribing a three-hour recorded interview with several speakers, a specialised tool will serve you better — and we would rather say so than sell you the wrong thing.

Transcriptor Light

Speech recognition and text generation both run on the iPhone. No account, no sign-up, no analytics, no telemetry. The app makes no network requests while you use it.

The practical shape of it: you finish a meeting, dictate a note in the corridor, tap once to turn it into a clean paragraph or a structured summary, and it stays on your phone.

One-time purchase, no subscription — which also means no vendor relationship to review, and no service that can change its terms next year.