Dictionary

Every speech model mishears the same category of word: proper nouns, product names,
internal jargon, and anything spelled unusually. The dictionary is where you fix that
once instead of every time. It lives under Vocabulary in the sidebar, in the
Dictation group.

It has two halves that work in completely different ways, and picking the right one
matters.

Vocabulary — spelling guidance

A vocabulary entry is a word you tell Zerm to expect. The list is passed to the model as
guidance: when these words, or words that sound like them, appear in your speech, spell
them exactly this way.

Use it for names and terms whose spelling is the problem — colleagues' names, product
names, libraries, medical or legal terms.

Vocabulary is guidance, not a rule. It shifts the odds; it does not guarantee. It is
used in two places:

• By the speech model, when the model supports it. Those models carry a
  Uses Dictionary badge — see models. Nothing needs switching on.
• By enhancement, as spelling guidance, whenever it runs.

Add several at once by separating them with commas. Duplicates are rejected rather than
silently added twice.

Word replacement — an exact rule

A replacement is a find-and-replace applied to the finished transcript, locally, before
the text reaches your cursor. It always fires, on every recording, in every output mode,
and in Transcribe File.

Use it when you know exactly what comes out and exactly what you want instead:

You say	You get
"my website link"	https://arcusis.github.io/Zerm/
"Voicing" or "Voice ink"	Zerm
"em dash"	—

How matching works

• Case-insensitive. "zerm", "Zerm", and "ZERM" all match.
• Whole words. Matches respect word boundaries, with punctuation counting as a
  boundary, so a replacement for "ink" does not corrupt "thinking". For scripts that do
  not space words, Zerm falls back to plain substring matching.
• Longest first. When several rules could apply, the longest original is tried
  first, so a specific phrase beats a shorter overlapping one.
• Several originals, one replacement. Separate the originals with commas — every
  variant maps to the same output. This is how you catch the four ways a model spells
  one name.

Each rule can be switched off without deleting it.

Which one to use

	Vocabulary	Word replacement
What it does	tells the model what to expect	rewrites the finished text
Certainty	improves the odds	always applies
Where it works	models marked Uses Dictionary, and enhancement	everywhere
Good for	names, jargon, unusual spellings	fixed phrases, expansions, known mistakes

If the model already produces something recognisable and you just want it spelled
differently, use a replacement. If the model is producing nonsense because it has never
heard the word, use vocabulary.

Quick add

There is a global shortcut for adding to the dictionary without opening the app — for
the moment you notice a mistake, rather than the moment you remember to fix it. See
shortcuts.

Import and export

Export Settings and Import Settings, in Settings, include the dictionary along
with your prompts, Power Modes, and custom models, so it moves with you between Macs.