Vicinae

Dictation

Vicinae can turn your voice into text and paste it into the active window. Transcription can run entirely on your machine, or through a cloud provider.

Setup

Dictation needs a transcription model. You can either:

  • Download a local model from the Built-in provider in the AI settings.
  • Add a provider that offers transcription.

Then select it in the Transcription model setting of the dictation extension.

On macOS, Vicinae will ask for microphone access the first time you dictate.

Dictating

Run the Dictate command to start recording. A HUD shows that recording is in progress. Run it again to stop and transcribe.

Dictate is best used with a global shortcut:

  • Press the shortcut to start, press it again to stop.
  • Hold the shortcut to talk, release it to stop (push-to-talk).

Once transcribed, the text is pasted to the active window, or copied to the clipboard, depending on the Dictation Action setting. If pasting fails, the text is copied to the clipboard instead.

Past dictations can be found in Dictation History.

Vocabulary

The Vocabulary command lets you list names and terms you use often, so that they are spelled correctly. Words can be added quickly with Add Word to Vocabulary.

Each entry can have an optional hint describing what it refers to. Hints are not used for transcription, but help the rewriting model (see styles) pick the right word.

Vocabulary is used to bias the transcription model towards the right spelling, but it is not a guarantee. Vicinae does not replace words based on similarity after the fact, as this proved too unreliable.

For best results, configure a style: the rewriting model receives your vocabulary and is much better at fixing misheard terms than the transcription model alone.

Styles

Styles let an AI model rewrite what you dictate before it is pasted: fix punctuation, remove filler words, change the tone, translate...

A style is made of:

  • A name
  • Instructions describing how to rewrite the text
  • The model to use for rewriting
  • The apps it applies to

An app belongs to at most one style, and a default style can be set to cover every app without its own. This makes it possible, for instance, to have a casual style for chat apps and a more formal one for your email client.

Styles are managed with Manage Dictation Styles and Create Dictation Style.

Dictation can be controlled from the command line, which is handy to bind it from your window manager or compositor:

vicinae dictation start   # start recording, no-op if already recording
vicinae dictation stop    # stop recording and transcribe
vicinae dictation toggle  # start, or stop and transcribe if recording
vicinae dictation cancel  # discard the current dictation

The same actions are available as deeplinks: vicinae://dictation/<start|stop|toggle|cancel>.