Can Dictation Learn My Names and Jargon?

Short answer: Yes, dictation can learn your names and jargon. A custom dictionary lets you add specific names, brands, acronyms and technical terms with the exact spelling you want. BlaBlaType applies that dictionary on your Mac, entirely on-device, so those words come out correct without you ever fixing them by hand.

Why standard dictation trips over your vocabulary

Speech recognition converts sound into the most statistically likely words. That works beautifully for everyday English, but your world is full of terms a general model has never weighted highly: a teammate named Siobhan, a product called Nexlify, an internal acronym like QBR, or a drug name that sounds like three other drug names. The model hears the audio correctly, it just guesses a common word instead of your specific one.

This is a known trade-off in speech recognition systems: broad accuracy versus niche precision. A general model cannot know that in your context, "cell" should be "Cel" or that "get" should be "Git." That gap is exactly what a custom dictionary closes. Instead of retraining anything, you hand the tool a short list of the words that matter to you and how they are spelled.

The model does not need a bigger brain. It needs your vocabulary. A custom dictionary hands it exactly the names and jargon you say every day.

How a custom dictionary actually works

A custom dictionary is a small list you control. You add a term, you set its exact spelling and casing, and from then on the app biases toward writing it that way when it hears the matching sound. On BlaBlaType this list lives on your Mac and is consulted during transcription, so there is no separate training step and no waiting.

Three steps get you there.

Because BlaBlaType runs its speech models locally, this whole loop happens on your machine. If you care about the mechanics of local processing, we cover it in how on-device AI rewrites text without the cloud, and the privacy side in is Mac dictation private.

Dictionary, AI cleanup, or say-it-out-loud?

People solve the "wrong word" problem in three ways. Only one of them scales.

Approach	Learns your terms	Effort per use	Stays on-device
Custom dictionary	Yes	Set once	Yes
On-device AI cleanup	Partly	Automatic	Yes
Spelling names out loud	No	Every time	Yes
Fixing by hand after	No	Every time	Yes

The dictionary is the durable fix: you invest a minute once and never spell that name aloud again. On-device AI cleanup is a strong partner because it keeps casing and formatting consistent, and a custom AI prompt can tell it how your domain writes things. Spelling terms out loud and correcting by hand both work, but they tax you on every single dictation, which defeats the point of talking instead of typing.

Do and do not: teaching dictation your words

A few habits make a custom dictionary far more reliable. Here is what pays off and what to avoid.

Do	Do not
Add the exact spelling and casing you want, such as "GitHub" or "PyTorch."	Assume the model will guess your preferred capitalization on its own.
Enter names for real people you mention often: clients, teammates, patients.	Wait until a report is due, then fix twenty misspelled names by hand.
Add acronyms and product codes that a general model would never predict.	Spell every acronym out loud on each dictation as a workaround.
Use a custom AI prompt to describe your domain so cleanup keeps context.	Rely on one tool to guess between two names that sound identical.
Review new jargon as it enters your work and add it in a few seconds.	Send your terminology to a cloud service just to get correct spelling.

If you find yourself narrating commas and periods to force structure, that is a separate habit worth dropping. See should I say punctuation out loud when dictating, because AI cleanup usually handles it for you.

Great for technical and specialized work

The value of a custom dictionary climbs with how specialized your language is. Developers dictate class names, CLI flags and library names that no general model expects, so the dictionary plus a domain-aware AI prompt keeps a term like Claude Code or "kubectl" intact instead of turning it into two ordinary words. Clinicians and lawyers get the same benefit for drug names, statute references and standard phrasing.

Once your vocabulary is loaded, dictation stops feeling like a compromise. That matters because most people speak around three to four times faster than they type, so the moment the tool stops misspelling your world, voice becomes the faster path for real work. If talking at your desk still feels odd, we address that in is talking to your computer weird at work. And if you want to see how the paid tiers unlock file transcription and screen-context features, the details are on pricing.