Where your voice actually goes when you dictate on a Mac
Jude Mercer, SEO and Content, Memoe — 14 October 2026
Dictation feels local. You talk, words appear. Whether your voice actually stays on your Mac while that happens depends entirely on which app you're using and how it's built — and for most of them, the honest answer is that it doesn't.
- Law firm Goodwin's own client alert on AI transcription tools puts it plainly: once audio and a transcript exist, they "often reside on systems controlled by the service provider, not the organization that initiated the recording," and depending on the vendor's terms, that data "may retain data indefinitely, share it with third parties, or use it for model training." The same alert flags that transcription tools which distinguish speakers can generate "voiceprints" — which Illinois's BIPA defines as a biometric identifier in its own right. (goodwinlaw.com, 27 April 2026; 740 ILCS 14/10)
- That's not a theoretical risk. On 18 December 2025, Katelin Cruz sued Fireflies.ai in federal court in Illinois, alleging the notetaker created and stored a voiceprint from her voice without ever telling her in writing that it was collecting biometric data, how long it would keep it, or giving her a chance to consent. (Fisher Phillips, "AI Meeting Tools Are The Latest Target of Illinois BIPA Class Actions")
- Even Apple's own Dictation isn't a blanket exception. Apple's own Mac Help says you can check Keyboard settings "to see whether your voice inputs and transcripts for general text Dictation … are processed on your device and not sent to Siri servers" — which language and region you're in decides the answer, and Apple leaves it to you to go and check. (Apple Mac Help, "Use Dictation on Mac")
Why dictation tends to leave the device at all
Turning speech into accurate text well, especially across accents, background noise and a user's own vocabulary, used to take more compute than a laptop could spare. The easiest way to get there was to stream the audio to a server built for exactly that and send text back. That's still how most dictation and meeting-notetaker products work, and it's why the server, not your Mac, ends up holding the recording, the transcript, and whatever metadata the vendor logs around it — the setup the Goodwin alert above is warning clients about.
How Moe's voice recognition works
Moe runs speech recognition and the voice it speaks back with on your Mac itself, using local models (NVIDIA Parakeet for listening, Kokoro for speaking) on the Neural Engine — not a server. Audio isn't stored once it's been turned into text, and on the free tier your voice never leaves your Mac at all.
The one opt-in exception, stated plainly: both paid plans include Cloud voice, a setting you can turn on for a faster, more natural-sounding voice. With it on, your Mac streams what you say to AssemblyAI to be turned into text, and sends the words Moe is about to speak to Inworld to be spoken aloud. Turn Cloud voice off in Settings and your voice goes back to being processed on your Mac only. Full detail, including what each of those two vendors is told and for how long: the Cloud voice section of the privacy page.
If you're checking a tool, not taking our word for it
The question worth asking about any dictation app isn't "does it say privacy-first" — it's where the audio and the transcript physically sit once the words appear on your screen, and whether that's disclosed somewhere more binding than a tagline. For Apple's own Dictation, the answer is in Keyboard settings, per Apple's guide above. For Moe, it's the privacy page, including the one feature that's an explicit, reversible exception rather than something left for you to discover later.