Vocabulary Correction

Speech engines often mishear names, product terms, acronyms, code words, legal phrases, and medical vocabulary. Vocabulary correction lets you define your own replacements after transcription.

What It Does

Vocabulary correction turns common mistakes into the terms you actually meant.

Examples:

Correct term	Common misrecognitions
Supabase	super base, soup a base
PyTorch	pie torch, pytorch
OpenRouter	open router, open rowder
ischemia	is key mia, iskemiya
force majeure	force major, forced measure

Dictionary Format

Use one correction per line:

Correct Term = misheard phrase, another misheard phrase
Supabase = super base, soup a base
PyTorch = pie torch, pytorch
OpenRouter = open router, open rowder

Keep entries specific. A correction that is too broad can replace text you did not intend to change.

How To Build A Useful Dictionary

Dictate normally

Do not try to guess every mistake in advance.

Notice repeated errors

Add only terms that the engine mishears more than once.

Add the correct term and variants

Put the intended term on the left and the mistaken phrases on the right.

Test with a short sentence

Make sure the correction works without breaking normal language.

Domain Dictionaries

Separate dictionaries are easier to control than one giant file.

Dictionary	Example entries
Technical	Supabase, PyTorch, LangChain, OpenRouter, Postgres
Medical	MedASR, ischemia, tachycardia, metformin
Legal	force majeure, indemnity, jurisdiction, plaintiff
Personal	client names, project names, internal acronyms

Turn domain dictionaries on only when they are relevant to the text you are dictating.

Vocabulary correction can use fuzzy matching to catch near-misses instead of only exact phrase matches. Use conservative thresholds for high-stakes terminology, because aggressive fuzzy matching can create false corrections.