Skip to main content
Accuracy depends on three things: the speech model, the words it knows, and the clean-up that runs afterwards. Pick a strong model, add the terms you use to your Dictionary, and turn on AI enhancement for punctuation and grammar. First check whether your model uses your Dictionary at all. See How vocabulary reaches each model.

Choose the right model

The model has the largest effect on accuracy. See Speech models for the full list.
  • Everyday dictation on a Mac: Parakeet Ultra, recommended during onboarding. Nemotron EN gives a punctuated live preview.
  • Everyday dictation on Windows: Parakeet V2 for English, Parakeet V3 for 25 European languages.
  • Languages outside Europe: Whisper Large v3 Turbo locally, Cohere Transcribe on a Mac, or a cloud model.
  • Fast or accented speech: cloud models usually cope better than small local ones.

Microphone tips

  • Use a dedicated microphone when you can. Earbuds beat a laptop microphone at arm’s length.
  • Dictate in a quiet room and keep a steady distance from the microphone.
  • Check that the right input is selected in Settings › Audio. See Microphones.

Speaking tips

  • Speak at a natural pace.
  • Pause briefly before and after speaking so the first and last words are not clipped.
  • Say technical terms and unusual names clearly.
  • Most models add punctuation themselves. Whether a spoken “comma” or “new paragraph” becomes punctuation depends on the model. With AI enhancement on, spoken punctuation commands are applied for you.

Use your Dictionary

Add the terms your model gets wrong: product names, jargon, names of people, acronyms.
1

Open Dictionary

Click Dictionary in the sidebar.
2

Add a word

Click Add Word and type the term as you want it spelled.
3

Add misspellings (optional)

List the ways the model mishears it, such as “cuber netties” for “Kubernetes”. See Misspellings.

How vocabulary reaches each model

Some engines steer recognition towards your terms. Some take them only as a hint. Some take nothing. These get nothing:
On Parakeet, Nemotron and Cohere, Dictionary words do not change what the model hears. Misspellings and Threads still work, because they run on the text afterwards. To influence recognition itself, use Whisper Small or larger, or a cloud model from the first table.
When your list is longer than a provider allows, Vowen keeps the terms you say most often, then the most recent.

Misspellings

A Dictionary word is a hint to the model. A misspelling is a fixed rewrite: an exact phrase that is replaced with the correct spelling after transcription, every time. Matching ignores case, so listing github under GitHub fixes capitalization on its own. Misspellings work with every model, including Parakeet, Nemotron and Cohere.

Auto-add corrections

Settings › General › Auto-add corrections to Dictionary watches what you fix by hand right after a dictation. When it spots a correction, it suggests adding it to your Dictionary. It is marked Beta, needs a Pro plan, and needs a connected AI provider. Settings › General rows for Save transcriptions to history, Remove filler words and Auto-add corrections to Dictionary (Beta) Vowen asks before changing anything. If you decline a suggestion, it is not offered again.

Pass screen as context

Names are hard for any speech model, and the right spelling is often already on screen, in a chat header or a document title. Pass screen as context sends the names visible in your current window to your AI provider so it can correct ones the model misheard. Find it in Language models › Configuration, under Enhance transcription with AI. It is macOS only, marked Beta, and off by default. It appears only when AI enhancement is on and a provider is connected.
  • It reads the focused window through Accessibility. It does not record the screen or take screenshots.
  • It runs only when AI enhancement runs, and the names go with the enhancement request you are already making.
  • If the reading is not ready when the transcript is, the dictation goes ahead without it.
This sends text from your current window to your AI provider. Turn it on only if you are comfortable with that.

Use Threads for long phrases

For text you dictate often, such as an address or a signature, add a Thread. Say a short trigger and Vowen inserts the full text. Open Dictionary › Threads and click Add Thread. See Threads.

Turn on AI enhancement

AI enhancement fixes spelling and grammar, adds punctuation and capitals, formats numbers, and applies spoken commands like “new paragraph”. Turn it on in Language models › Configuration › Enhance transcription with AI.

Remove filler words

Settings › General › Remove filler words strips fillers before the text reaches AI enhancement. It is on by default and works without an AI provider. The default list is: uh, um, uhm, umm, uhh, uhhh, hmm, mmm, mh, ehh, äh, ähm, öhm, hm. ElevenLabs Scribe v2 can also remove fillers on its side. See ElevenLabs.

Set the dictation language

When you speak one language, set it instead of relying on auto-detect. Open Settings › Language › Dictation Language and click Change. Some models, such as Parakeet, only auto-detect, and the dialog says so. The Dictation Language dialog with Parakeet active, noting that Parakeet only supports auto-detect For British spellings, set Settings › Language › English Spelling to British English instead of writing an AI instruction. See Languages.

Back up your Dictionary

Your Dictionary and Threads are included in Settings › Account › Import / Export Data, a Pro feature. Use it to move your terms between computers or keep a backup.