Speed for long sessions
- No uploads or downloads—performance stays steady on hotel or café Wi-Fi.
- Voice Type streams in ~30 second windows. When you stop dictation, only the last window finalises (typically ≈2–3 seconds on an M1 Mac).
- Cloud tools often round-trip larger, uncompressed audio for best accuracy. Short bursts feel fine; longer recordings slow down.
Accuracy you can feel
- Audio is normalised to consistent loudness and cleaned before recognition, so transcripts need less editing.
- Noise-aware voice detection avoids importing keyboard clicks and room chatter.
- We improve the signal before recognition instead of "forcing" fixes with heavy prompts that can change what you said.
Privacy and control
- On-device by default—audio never leaves your Mac.
- Mac App Store distribution with Apple sandboxing and notarisation.
- Optional bring-your-own-key rewrites let you decide if and how text is sent to a provider.
When cloud can still make sense
- Team workflows that require server-side storage or multi-device sync.
- Specialised medical or legal language models hosted by a provider.
- Environments with consistently strong connectivity where managed infrastructure is preferable.
Who benefits most from offline dictation
Writers, researchers, clinicians, students, and engineers who dictate frequently gain the most. If you value privacy, consistent speed, and transcripts that match what you actually said, offline dictation is the best fit for macOS.