everyday mac tools

Where your audio goes when you dictate on a Mac

· 6 min read

It depends on the machine and on the app, and the honest answer is that you should check rather than assume. Keyboard dictation on an Apple silicon Mac processes many languages on the device; on older Intel hardware, macOS sends the audio to Apple for processing. Most dictation apps that are not Apple’s send audio to a service unless they specifically say otherwise, so the question to ask about any of them is what happens by default and what you can change.

The three places your voice can end up

Every dictation product sits in one of three positions, and the marketing rarely distinguishes them.

On the device. Speech is turned into text by the Mac itself. Nothing leaves, which also means it works with the network off, and the recognition model is limited to what fits and runs locally.

Sent, processed, discarded. Audio travels over an encrypted connection to a server, comes back as text, and the recording is thrown away. This is the common arrangement and it is usually the most accurate, because the model on the other end is larger than anything that would run on your laptop.

Sent and kept. Audio is stored, sometimes to improve the service, sometimes because storage was easier than deletion. Where recordings are retained, there is usually a setting or a review program behind it, and it is the position most people assume they are in when they are not.

None of the three is wrong in the abstract. What matters is knowing which one you are in before you dictate something you would not put in an email.

What macOS Dictation actually does

Open System Settings, Keyboard, and look at Dictation. Next to it is a note headed About Dictation and Privacy. Read it once on your own machine, because it describes the behavior for your hardware and your language rather than for Macs in general.

The broad shape: on Apple silicon, keyboard dictation in supported languages is processed on the Mac, which is why it keeps working when the Wi-Fi is off. On an Intel Mac, and for some languages or for longer requests, audio is sent to Apple for processing instead. If it matters to you, test it rather than trusting a summary, including this one. Offline dictation on a Mac has the test.

Two related settings live elsewhere in System Settings and are worth a minute.

  1. In Privacy and Security, open Analytics and Improvements. There is a control that governs whether Apple may keep and review audio samples from Siri and Dictation. If you have never looked at it, look now and set it deliberately.
  2. In the Siri section of System Settings there is a button that deletes the Siri and Dictation history associated with your device. Use it if you have been dictating for years without thinking about it.

Voice Control, in Accessibility, is a separate feature with its own engine and its own downloads. It is designed to run on the device, and it does both dictation and command of the interface. It is heavier to set up than keyboard dictation and better if you are moving away from the keyboard entirely.

Seeing when a microphone is open

macOS gives you a live indicator, and it is more useful than any policy document.

When any app is using the microphone, an orange dot appears at the right of the menu bar. Click Control Center and the top of the panel names the app that is listening. If you ever see the dot when you are not dictating, that is the fastest way to find out who is responsible.

The list of what is permitted lives in System Settings, Privacy and Security, Microphone. Every app that has asked for the microphone is there with a switch. Turn off the ones you do not recognize, and be aware that apps which record in the background have to appear in this list; there is no way around it for an app distributed normally.

What to ask about any dictation app

Five questions, and a product that answers them clearly is telling you something in itself.

  1. Does audio leave the device in the default configuration? Not “can it stay local”, but what happens if I change nothing.
  2. Is the audio stored after transcription, and for how long?
  3. Is there a mode that keeps everything on the machine, and what do I give up by using it?
  4. What happens to the text? A dictation history is convenient and it is also a transcript of everything you said, sitting somewhere.
  5. Who does the processing? If a third party does the transcription, that is a second organization with your voice in it.

For our own app the answers are these. Voice Keyboard Pro sends audio, by default, over an encrypted connection to a transcription provider, and the audio is discarded after transcription rather than stored. That is the default, not an accident, and it is the arrangement that gives the accuracy. Its Settings also contain an Offline Mode that runs transcription entirely on the device with no network at all. It is not offline unless you turn that on, which is the sort of sentence a product page tends to leave out.

Practical settings if you dictate sensitive things

If your dictation involves clients, patients, legal matters or anything under an agreement you signed, decide this before the first session and not after.

  • Turn on on-device processing where the app offers it, and confirm it by dictating with the network off.
  • Keep sensitive dictation out of browser text fields where a page’s own scripts and extensions can see what appears.
  • Know where the dictation history lives, and clear it on the same schedule you clear anything else.
  • Check the microphone list in Privacy and Security occasionally, particularly after installing anything that joins meetings.
  • If you are covered by a professional obligation, the decision about which mode is acceptable belongs with whoever handles that where you work, not with you alone at your desk.

Two of the studio’s guides go further into specific settings: dictating clinical notes on a Mac and case notes by voice for lawyers on a Mac.

Questions

Is dictation on a Mac private? Keyboard dictation on Apple silicon in a supported language is processed on the Mac, which is about as private as this gets. Other combinations send audio to Apple. Neither statement covers apps that are not Apple’s, and those vary from fully local to fully hosted.

Does turning off Wi-Fi prove nothing was sent? It proves nothing was sent while it was off, which is a genuinely useful test and the only one you can run yourself. It does not prove anything about the app’s behavior when the network returns, so use it alongside the app’s own description rather than instead of it.

Is on-device recognition worse? Usually a little, and most obviously on names, jargon and unusual words, because the local model is smaller. A personal vocabulary, where the app has one, closes a lot of that gap. For everyday prose the difference is smaller than people expect.

What about dictation on an iPhone? The same three positions apply, and recent iPhones process a good deal of keyboard dictation on the device. As on the Mac, an app you install is a separate question from what iOS itself does, and the app’s own settings are where the answer is.