VoicedDictate anywhere your Mac can type.
Voiced records while you hold a key, transcribes locally with WhisperKit, then copies or pastes clean text back into your Mac.
Voiced
Menu bar capture
Output mode
Paste, then restore clipboard
Model
WhisperKit tiny, verified
"Draft the release note. Mention macOS only, local transcription, and that copy-only mode is available."
Transcribed locally, ready for the focused app.
Dictation that stays out of the way.
The whole loop is designed for the moment between thinking and typing: hold a key, speak, release, then let Voiced put the text where you need it.
Hold the hotkey
Press and hold Right Command, or choose another push-to-talk key in settings.
Speak in place
A small indicator follows the recording state, with optional sounds for start and stop.
Paste or copy
Voiced can paste into the focused app or leave the transcript on your clipboard.
Floating indicator
Recording, transcribing, model loading, and error states stay visible without opening a window.
Focused-app output
Paste mode sends Cmd+V with Accessibility permission. Copy-only mode skips synthetic keystrokes.
Local transcription without pretending setup is invisible.
Voiced uses WhisperKit and Argmax OSS models on your Mac. The first approved download can contact Hugging Face, then the model is used from the local cache.
Models are downloaded deliberately.
Voiced asks before loading the local transcription model. It shows progress, verifies downloaded files against pinned hashes, and exposes model storage controls in settings.
- Explicit model download approval
- SHA-256 integrity verification
- Local WhisperKit transcription
- Delete or reveal downloaded models
Capture has visible limits.
After setup, speech is recorded locally, transcribed locally, then emitted as text. Temporary audio is cleaned up, and transcript content is not stored as history.
- No cloud transcription path
- Temporary audio deleted after transcription
- Last transcript recovery is memory-only
- Copy-only mode avoids Accessibility paste
A small tool with explicit boundaries.
Voiced is built around a simple bargain: record only while push-to-talk is active, transcribe locally after model setup, and keep the output path under your control.
Mac utility, not a web account
Runs from the menu bar with macOS permissions you can see and change.
Plain privacy boundaries
Privacy, diagnostics, model downloads, and permissions are explained before they matter.
Transcript text stays out of logs
Logs use counts and state changes, not audio content, transcript text, file names, or window titles.
Ready when your model is