How vocabulary works#
Evoglyph works in two steps. The speech model writes down what it heard, then the cleanup model turns that raw text into finished text. Your vocabulary list guides the second step.
Before cleanup runs, Evoglyph compares the raw text with your terms. When part of it comes
close to one of them, such as "post gress" for Postgres or
"oh auth" for OAuth, Evoglyph hands that term to cleanup. Cleanup
then spells it the way you wrote it.
Only terms that come close to what you said reach cleanup, so a long list does not crowd
the model, and a term you never said does not slip in. Evoglyph also skips a match when
the words it would replace are all ordinary English, so a term like Go does
not take over the verb "go". Cleanup still makes the final call, so a term that sounds
very different from how it is spelled may come through as heard.
There's no model retraining, no per-user fine-tuning, and no audio leaving your Mac. Adding or removing a term takes effect on the next dictation.
Earlier versions also ran a second speech model that swapped vocabulary terms into the raw text. Version 0.3.8 turned it off, because it added wrong terms more often than it fixed missed ones.
How to add, edit, and remove terms#
Vocabulary is managed in the Vocabulary tab of the Evoglyph dashboard. For a full overview of the dashboard and its other tabs, see App dashboard.
Open Evoglyph's dashboard from the menu bar icon, then go to Vocabulary. The vocabulary panel is a simple list: one term per row, an Add button at the top, and a delete control on each row.
- Add: type the term exactly as you want it transcribed (capitalization included), press Enter or click Add.
- Edit: click the term to edit it in place. Save with Enter, cancel with Escape.
- Delete: click the trash icon on the row.
Changes save immediately. The next time you dictate, your updated list is in effect; there's no apply step, no restart.
The vocabulary UI is being polished as part of the launch onboarding work (Evoglyph#697). The behavior described above is stable; the visual layout may shift slightly between releases.
What to add (and what not to)#
Good candidates are words the base model has no reason to know: your domain's jargon, brand names, and acronyms. Some categories that pay off immediately:
-
Software acronyms:
gRPC,OAuth,Postgres,JSON,YAML,WebSocket,WASM -
Proper nouns:
Evoglyph,Cloudflare,Anthropic, your company name, your teammates' names -
Command-line tools:
pnpm,wrangler,gh,kubectl,tmux,jq,rg -
Frameworks and libraries:
Tailwind,SwiftUI,FluidAudio,vitest, project codenames -
Words you say often that get mistranscribed: if Evoglyph keeps writing
"graph QL" when you said "GraphQL", add
GraphQLand move on
Bad candidates: common English words. Adding the, and,
but, that doesn't help. The base model already knows them, and
Evoglyph skips matches made only of ordinary English words, so these entries never reach
cleanup.
Best practices#
- Keep the list focused. A list of the words you actually say works better than a dump of every technical word you've ever read. More terms means more chances for one to match a similar word you did not mean.
-
Capitalize as you want it injected. Evoglyph hands cleanup your term
exactly as you typed it. If you add
Postgres, cleanup aims forPostgresin your output, notpostgres. -
Record samples for terms that sound unlike their spelling. Evoglyph
matches the letters of the raw text, not the sound. If you say "ess three" for
S3, record a few pronunciation samples when you add the term. Evoglyph saves how it heard you, so later dictations can match that form. - Prune occasionally. If a term is no longer relevant (an old project, a tool you stopped using), delete it. A stale term can match a similar word you actually meant.
-
Don't add multi-word phrases unless necessary. Matching works best on
single words and short compounds.
OpenSlideAIis fine; "the quarterly business review process" isn't a vocabulary entry, it's a sentence.
Troubleshooting#
Open the Vocabulary panel and confirm the spelling is exactly what you expect. Cleanup
uses the term as written, so Postgress comes out as Postgress.
If the spelling is right, the speech model may be hearing something too far from the
term to match. Record a few pronunciation samples for it so Evoglyph learns how it hears
you.
Cleanup uses the casing of your entry. Edit the entry to match the casing you want. If
you have both Postgres and postgres in the list, delete the
one you don't want.
Check that you're adding the right kind of term. Vocabulary helps most with rare or
domain-specific words. Common English (that, about,
over) never reaches cleanup as a vocabulary match. Restrict your list to
words the base model is unlikely to have seen often.
The term is close enough to a word you use for something else that cleanup swaps it in. Delete it, or make the entry more specific. A short entry is the usual cause; replace it with the full term you mean.
For deeper accuracy issues unrelated to vocabulary (slow transcription, wrong audio device, or empty output), see the troubleshooting guide.