16 JULY 20263 MIN READPRIVACY

Cloud vs local dictation: where does your voice actually go?

A plain-English map of the data path behind every dictation app — what leaves your machine, who can hear it, and the questions to ask before you install one.

You press a hotkey, you talk, text appears. Between those two moments, your voice takes a journey — and depending on the app, that journey is either a few centimetres of silicon or a round trip through several companies' infrastructure.

This post is a map. It applies to any dictation tool, not just ours.

The three architectures

1. Cloud-processed

The dominant model. Audio is captured locally, then uploaded to the vendor's servers, where their speech models transcribe it and (usually) their language models clean it up. This is how most modern AI dictation apps work, typically sold as a monthly subscription.

What to understand about it:

  • Your audio and your text transit and are processed on infrastructure you don't control. How long they're retained, whether they're used to improve models, and who inside the company can access them are policy questions — read the specific vendor's answers.
  • The vendor sits between you and the model, so pricing is per-seat rental, and the app stops working if the company or your subscription does.
  • The upside is real: server-class models, no local compute needed, works on any device.

2. Local / on-device

Everything happens on your machine: speech-to-text runs on your own processor, and so does the clean-up model. Nothing about the content of your dictation touches a network.

  • Nobody — including the vendor — can access what you dictate. Not because they promise not to, but because they never receive it.
  • Works offline; latency doesn't depend on your connection.
  • Requires capable hardware. On Macs this effectively means Apple Silicon, whose Neural Engine and GPU handle both stages comfortably.

3. Bring your own key (BYOK)

The middle path. The app is a client, and you supply an API key for a model provider you already trust. Audio or text goes directly from your machine to that provider — the app vendor is not in the data path and takes no cut of usage.

  • You inherit the provider's privacy terms — but you chose the provider, you hold the key, and you can revoke it any time.
  • You pay the provider's actual usage costs, not a marked-up subscription.

Questions to ask any dictation app

  1. Does my audio leave my machine — and in which modes? "We take privacy seriously" is not an answer; a data-path diagram is.
  2. Whose servers process it? The vendor's? A cloud AI provider's? Both?
  3. What exactly does the vendor's own backend see? Even local-first apps usually have licensing or update services — those should carry no content.
  4. Where do API credentials live? On macOS the right answer is the Keychain, not a plaintext config file.
  5. What happens offline? Offline operation is the easiest proof that processing is genuinely local.

How ShoutFlow answers them

We built ShoutFlow so the answers are structural rather than contractual: local transcription (WhisperKit on the Neural Engine) and local clean-up (MLX on the GPU) by default on Apple Silicon; optional BYOK providers that your Mac calls directly; keys in the macOS Keychain; and a licensing service that sees your license status and nothing else. The full detail is in our privacy policy — including what our servers do see, because "nothing at all" would be a red flag, not a feature.

Whichever tool you pick, pick it knowing where your voice goes.

← ALL POSTSGet ShoutFlow — $25