How to fix misheard jargon in dictation with a vocabulary list

Why does dictation get technical words wrong?

A speech engine writes the ordinary words it knows best, so a rare name loses to a common word that sounds alike. It has met words like report and meeting far more often than your team’s tool names and acronyms, and when the sound is close, the familiar word wins.

Say your team works in a tool called Fennick. The engine may type fennel, or split the name into two short words that sound right and mean nothing. An acronym said letter by letter, such as JQV, can come out with spaces or full stops between the letters.

What does a vocabulary list do?

A vocabulary list is the set of names and terms you add on the Vocabulary page of Dictate Safe, a dictation app, which uses them to spell what it hears your way. Type the term exactly as it should be spelled and press Add, and the next dictation uses it.

• Your terms are given to the speech model before it listens, which raises the odds that it picks your word over one that only sounds like it.
• A term heard with the right letters but the wrong capitals or spacing is put right, with no AI involved: “fennick ops” comes out as FennickOps if that is how you saved it.
• AI cleanup, if you switch it on under Settings → Cleanup, also corrects a listed term that came out close but wrong. It never rewords what you said.

Dictate Safe dictates in English only, so the list is for English dictation and the names you use in it.

How do you write a vocabulary list that works?

Add only the terms that keep going wrong, and type each one exactly as you want it typed. We think a short list of words that matter does more good than a long one of words you might say one day.

• Wait until a term has gone wrong more than once. A word that is wrong once might be noise.
• Type the capitals and spacing you want, such as JQV or FennickOps, because the app copies your spelling. The page will not take a second copy of a term that differs only in capitals, so to change one, remove it with its cross and add it again.
• Put one term in each entry. The box saves whatever you type as a single entry, so names separated by commas become one long term that will rarely match anything you say.
• Leave out everyday words. The engine already knows them.
• Prune it. Every term makes the engine more ready to write it, so one you never say can turn up in place of a word that sounds like it. The page shows a count beside each term once it has been typed, and one with no count after a few weeks can go.

Why does a name still come out as a common word?

A list raises the odds of the right word without guaranteeing it, so a rare name that sounds exactly like an everyday word can still come out as that word. If your team’s tool is called Nyte and you say it just like night, the engine has every reason to type night. The spelling fix only changes capitals and spacing, and AI cleanup will not always catch it.

Saying the name inside a longer phrase, such as “the Nyte dashboard”, can help, because the words around it give the engine more to go on.

A snippet is the surer route. On the Snippets page, save a trigger phrase of two ordinary words you would never say by accident, such as night app, and set it to expand to Nyte. Saying night app then types Nyte exactly as saved, and no AI touches it.

How can you check your list in five minutes?

Import the same two short audio files before and after you add your terms, so the list is the only thing that changes.

1. Write one sentence that uses each problem term once, and a second that uses none of them.
2. Say each sentence into its own audio file, for example with Voice Memos on a Mac. Import reads M4A, MP3, WAV and FLAC.
3. With Code Mode off and the terms not on your list yet, open Import in the sidebar, drop both files on it, and copy the two texts.
4. Add the terms on the Vocabulary page, change nothing else, and import the same files again.
5. Compare. The first sentence shows what the list fixed, and the second whether it changed words you never listed.

Import uses your list and the same speech engine as your dictation. If a term now has the wrong capitals, check how its entry is typed; if it is still an everyday word, try a snippet. The steps for Import are in its help article.