Vocabulary
Speech recognition can misspell names, acronyms, and technical terms. Vocabulary lets you teach Stirae the spellings you prefer so normal dictation can use them more consistently.
Real examples
| Speech recognition hears | Stirae outputs |
|---|---|
| "kuber netes" | Kubernetes |
| "quartz queue el" | QuartzQL |
| "AWS" | AWS (kept untranslated) |
| "mahakavi" | Mahakavi (exact spelling) |
Adding a term
Open Settings > Vocabulary
Click New Vocabulary.
Enter the Written Form
Exactly how you want it written: "Kubernetes", "macOS", "AWS".
Add spoken forms
Variations that speech recognition might hear. For Kubernetes: "kuber netes", "kuber netties".
Set behavior flags (optional)
- Preserve spelling — prevents refinement from changing casing (e.g., keeping "macOS" not "MacOS")
- Do not translate — keeps the term unchanged even when translating output
- Use With — limit to a specific spoken language, or leave as "Any"
Managing terms
Use Search vocabulary to find an existing term. Click a term row to edit it, or use the row's more menu to choose Edit… or Remove….
If Stirae cannot save a Vocabulary change, it keeps using your previously saved terms.
How it works during dictation
For normal dictation runs, Stirae uses your saved Vocabulary in three places:
- After transcription — spoken forms are swapped for the written form before later steps see the text
- During refinement or translation — Stirae includes your active terms when the current cloud path supports vocabulary guidance
- Before insertion — a final pass restores preserved terms and terms marked Do not translate if a later step changed them
If a Voice Shortcut matches, Stirae inserts the shortcut replacement text and skips the final Vocabulary pass for that run.