Word replacements
Word replacements rewrite text after recognition finishes — for fixed errors, brand spellings, and filler words.
- Settings → Text processing → Word replacements
- Menu bar entry
- Deeplink:
voicehotkey://word-replacements(see Deeplinks)
Rule fields
Section titled “Rule fields”| Field | Meaning |
|---|---|
| Original | Match text; comma-separated variants OK |
| Replacement | Result (empty = delete) |
| Regex | Optional; $1 groups |
| Timing | Before processing / After / Both (relative to LLM polish & translation; if no AI runs, rules still apply once to the raw transcript) |
Matching is usually case-insensitive; literal rules match on word boundaries where possible. For regex, add \b yourself.
Examples
Section titled “Examples”| Original | Replacement |
|---|---|
raycast, reycast |
Raycast |
air pods |
AirPods |
\b(um|uh)\b (regex) |
(empty) |
Difference from recognition dictionaries
Section titled “Difference from recognition dictionaries”- Engine context / prompt: helps ASR hear it right the first time.
- Word replacements: an output-level safety net, works with every engine.
You can use both together.
