Custom Vocabulary in Dictation: How Dragon, Apple, Wispr Flow and Others Handle It

Dictation gets ordinary English mostly right. Where it falls over is the words that matter most to you: your surname, your clients, your product, your field's acronyms. Custom vocabulary dictation is how you fix that, and every tool handles it differently, from a 1990s-style vocabulary editor to an AI dictionary that learns from your corrections.

This guide explains how custom words work under the hood, then what each major tool offers, from each vendor's own documentation as of October 2026.

Why dictation mishears your words#

A speech recognition model predicts the most likely words for the sounds it hears. "Likely" comes from what it saw in training. Common words have huge weight; your colleague "Siobhan", your product "Kestrl" or the drug "apixaban" have little or none. So the model picks something that sounds similar and is more common: "Shivon", "kestrel", "a pixie ban" (made-up examples, but the pattern is familiar).

Custom vocabulary shifts those odds, or cleans up afterwards.

Two ways custom words work#

1. Hints to the recognizer#

The tool passes your words to speech recognition along with the audio, so the model is more likely to choose them when the sound fits. Different engines do this differently: some boost specific words, some take a short text prompt with the words in it. Either way, the fix happens during recognition, so it can handle words it hasn't seen you say before, and endings like plurals.

The catch: hints are nudges, not rules. A word can still be missed, and a very long list can backfire by pulling the model toward the wrong words.

2. Replacements after recognition#

The tool looks for a known wrong output and swaps in the right one: "kestrel" becomes "Kestrl", "lume voice" becomes "LumeVoice". This is completely predictable and great for capitalization and odd spellings.

The catch: it only fixes mistakes you've already seen, and it can misfire if the wrong form is also a real word you sometimes mean ("kestrel" the bird).

Older systems such as Dragon also learn pronunciation: you say the word, and the engine stores how you say it. Most cloud AI dictation tools don't ask for this.

The practical answer for most people is both: hints for the names, replacements for the stubborn ones.

Dragon: the vocabulary editor#

Dragon has the most detailed controls of any tool here, which makes sense for a product long used in medicine and law. According to Nuance's Dragon help on adding words:

• Vocabulary Editor (DragonBar > Tools > Vocabulary Center): add, edit or delete words and phrases, give a word a different spoken form, and optionally train its pronunciation.
• Learn from specific documents: point Dragon at your own files and it picks up the words and phrases you use.
• Scope: custom words are tied to your user profile on that computer, unless an administrator sets up central management.

The help pages we found are for Dragon 15, so menu names may differ in your version. Dragon is Windows-only today; see Dragon pricing and LumeVoice vs Dragon.

Apple: none in Dictation, some in Voice Control#

Apple Dictation has no custom vocabulary. On Mac, iPhone and iPad there's no list where you can add a word for dictation to learn. Text Replacement in Keyboard settings expands shortcuts you type; it isn't a dictation vocabulary.

Apple's Voice Control is different. It's an accessibility feature for controlling your device entirely by voice, and it has a Vocabulary setting. Apple's guide, Use a custom vocabulary with Voice Control on Mac, says you can:

• Add up to 1,000 terms in System Settings > Accessibility > Voice Control > Vocabulary.
• Select a word on screen and say "Add to vocabulary".
• Import a text file with one term per line, and export the list.
• Not every language supports it; the option is dimmed if yours doesn't.

Voice Control changes how your whole Mac responds to speech, so it's a big switch just to get names right. For most people who simply want dictation to spell things correctly, that's where a dictation app comes in. See Apple Dictation alternatives.

Windows: no custom words in voice typing#

Windows voice typing (Windows key + H) has no custom vocabulary as of October 2026. On Copilot+ PCs, Fluid dictation tidies grammar and filler words in English, but there's still nowhere to teach it a name. The older Windows Speech Recognition had a speech dictionary, but Microsoft has deprecated it in favor of voice access. More in our Windows voice typing guide.

Wispr Flow: a dictionary that learns#

Wispr Flow has a Dictionary. Its help page describes two ways in:

• Manual: add industry terms and unusual spellings in the Dictionary tab.
• Auto-learned: when you correct the spelling of a single word after dictating and press Enter, Flow can add it automatically, shown under an "Auto-learned" label.

Auto-learning is a real convenience: you fix a name once and it tends to stick. The help page is dated, so the details in your version may differ. See LumeVoice vs Wispr Flow.

Superwhisper: vocabulary and replacements, separately#

Superwhisper splits the two approaches cleanly, which makes it a good illustration of the theory above. From its documentation:

• Vocabulary words are hints sent to the transcription model. Superwhisper's vocabulary docs advise using them sparingly, because too many words can confuse the model.
• Replacements run after transcription. They're applied programmatically, not by AI, so results are predictable. Matching ignores case, and the output uses the case you set.

Superwhisper can also run models on your Mac, which is its big advantage for privacy-sensitive work. See LumeVoice vs Superwhisper.

LumeVoice: a dictionary on every device, strongest on desktop#

LumeVoice has a custom dictionary, and how it works depends on the platform:

• Mac and Windows: names, jargon and acronyms you add are passed to speech recognition, so they come out spelled right in the first place. That's the "hints" approach.
• iPhone and Android: the dictionary's corrections apply. If a word keeps getting misheard, set the right spelling and LumeVoice fixes it in the text. That's the "replacements" approach.

So if your vocabulary is heavy (a medical practice, a law firm, a dev team with a dozen product names), do most of your jargon-heavy dictation on a computer, and use corrections to catch the stubborn words on your phone. The dictionary sits alongside LumeVoice's other cleanup: filler words removed, self-corrections applied, punctuation added from context, and snippets that type a long phrase from a short spoken cue. LumeVoice is cloud-based and needs an internet connection; if your vocabulary includes material that must stay on your computer, a local tool such as Superwhisper may suit you better.

At a glance#

Details as of October 2026, from each vendor's documentation.

How to build a vocabulary that works#

1. Start from real mistakes. For a week, note every word dictation gets wrong. That's your list.
2. Add proper nouns first. People, companies, products and places cause most errors.
3. Write words exactly as they should appear, with the capitals: "iOS", "McAllister", "SaaS".
4. Keep it short. Tens of words, not thousands, unless your tool is built for big lists like Dragon.
5. Add a replacement for the stubborn ones. If a word is still misheard after you add it, map the wrong output to the right one.
6. Avoid replacements that are real words you also use, or you'll "fix" the wrong thing.
7. Speak clearly and naturally. A name said in a full sentence gives the model context; a name said alone gives it none.
8. Review it every few months. Drop old clients and projects; add new ones.

Related reading#

• How to fix poor dictation accuracy on Mac
• How audio to text converters work
• Best dictation software