Short answer: The fix for a dictation app that keeps mishearing the same words is a custom vocabulary, also called a personal dictionary. You add the names, terms, and acronyms it stumbles on, optionally with replacement rules, so the app substitutes the correct spelling every time instead of guessing.
You have probably had this exact moment. You dictate the same client's name for the tenth time this week, and for the tenth time the app spells it three different wrong ways. Or you say a product name, a drug name, a codebase term, a colleague's surname, and it confidently types something that sounds vaguely similar but is wrong. You correct it. You move on. Tomorrow it happens again.
The frustrating part is that the mistake is so consistent. It is not random noise. The app gets the same word wrong in the same way, which feels like it should be the easiest thing in the world to fix. And it is, if your dictation app has the right feature. The feature you are looking for is usually called a custom vocabulary or a personal dictionary, and once you understand how it works, those repeat mistakes mostly disappear.
Why your dictation app keeps getting the same words wrong
To fix the problem, it helps to understand why it happens in the first place. Speech recognition does not hear letters. It hears sound, and then it has to decide which words those sounds most likely represent. For common English words this is easy. There are not many things "the meeting is on Tuesday" could plausibly be.
The trouble starts with words that are rare, invented, or specific to your world. Consider what a transcription engine is up against when you say a surname like "Nguyen," a startup called "Klyro," an internal tool named "Pegasus-2," or an acronym like "EBITDA." These are not in everyday speech, so the engine falls back on whatever common words sound closest. "Klyro" becomes "Cleero" or "clear oh." "Nguyen" becomes "win" or "new yen." The engine is not broken. It is doing exactly what it was designed to do, which is pick the most statistically likely interpretation of a sound, with no idea that this particular name matters to you.
This is the same root cause behind a lot of accuracy complaints. We wrote a whole separate guide on why Mac dictation types inaccurate or wrong words, and the pattern is almost always the same: the engine is guessing at vocabulary it has no reason to know. The difference between a frustrating dictation experience and a great one often comes down to whether you have told the app about your specific words.
What "learning" actually means here
When people search for a dictation app that "learns" the words it keeps getting wrong, they usually imagine something that silently absorbs corrections over time, like autocorrect on a phone. That kind of passive learning exists in some tools, but it is slow, unpredictable, and you have no control over it. It might pick up your correction, or it might not, and you cannot see what it has learned or fix it when it learns the wrong thing.
A custom vocabulary is the opposite, and it is far more reliable. Instead of hoping the app eventually figures out your terms, you tell it directly. You add the word once, and from that point on the engine knows it is a real candidate. You are not waiting for the app to learn. You are teaching it, explicitly, in seconds. For the words that matter most to your work, that explicit control is exactly what you want.
There are two flavors of this, and good dictation apps offer both:
- Vocabulary hints. You provide a list of terms that the engine should treat as known words. When a sound is ambiguous, your terms get weighted as likely candidates, so "Klyro" wins over "clear oh."
- Replacement rules. You define an exact substitution: whenever the engine produces a certain phrase, swap it for the correct text. This is the heavy artillery for words that are spelled nothing like they sound, or for shorthand you want to expand.
The fix: build a custom vocabulary that actually sticks
Here is the practical workflow. It takes about ten minutes the first time and then a few seconds whenever a new problem word shows up.
1. Make a list of your repeat offenders
For one normal day of dictation, keep a quick note of every word the app gets wrong more than once. Do not overthink it. You are looking for the names, jargon, brands, and acronyms that come up in your actual work. Most people are surprised that the list is short. It is usually twenty to forty words that account for the vast majority of their frustration. A salesperson might have a dozen client and product names. A developer might have a handful of repo and service names. A clinician will have drug names and abbreviations.
2. Add them to your custom vocabulary
Open your dictation app's vocabulary or personal dictionary settings and enter each term. Spell it exactly the way you want it to appear, including capitalization. If the app supports it, this is where you also add the proper-noun versions of names so they come out capitalized instead of lowercase.
3. Add replacement rules for the stubborn ones
Some words resist plain vocabulary hints because they sound exactly like a common word or phrase. For those, a replacement rule is the answer. If the engine reliably hears your company "Anoma" as "an oma," you create a rule: replace "an oma" with "Anoma." Now it is fixed every time, deterministically, regardless of context.
4. Use replacement rules for shorthand and expansions
Replacement rules are not only for fixing errors. They are a productivity feature in disguise. You can dictate a short trigger phrase and have it expand into something long you do not want to say in full: an email signature, a legal disclaimer, a standard closing, a formatted address, or a tricky technical string. Say the trigger, get the full text. This is one of the most underused tricks in voice dictation.
Words worth adding first
If you want the biggest accuracy jump for the least effort, prioritize these categories:
- People's names. Colleagues, clients, and contacts, especially surnames and any name that is not common in English. This single category fixes a huge share of everyday errors.
- Company and product names. Your employer, your products, the vendors and tools you mention constantly. Invented brand names are almost always misheard until you add them.
- Industry jargon and acronyms. The terms of art in your field. A finance professional needs EBITDA and ARR; an engineer needs API, OAuth, and their service names; a marketer needs their campaign and channel names.
- Place names. Streets, neighborhoods, and towns that are local to you but obscure to everyone else.
- Anything you have corrected twice. The simplest rule of all. If you have fixed a word by hand more than once, it belongs in your vocabulary.
Specialized fields benefit the most from this. We have seen exactly how much it matters in technical domains, which is why we wrote a dedicated piece on medical dictation for drug names and ICD-10 codes, where the entire workflow lives or dies on whether the app knows the vocabulary. The principle is identical whether your hard words are medications, statutes, or microservices.
How Voice Keyboard Pro handles this
This is exactly the problem Smart Vocabulary in Voice Keyboard Pro is built to solve. Voice Keyboard Pro is a voice-to-text app for Mac and iPhone, and on the Mac you hold a hotkey, speak, and release, and the text appears at your cursor in whatever app you are using. The Smart Vocabulary feature is your personal dictionary: a place to add the names, terms, and acronyms specific to you, plus replacement rules that swap a phrase the engine produces for the exact text you want.
Because the vocabulary feeds directly into Voice Keyboard Pro's transcription engine, your terms become real candidates rather than long shots. The replacement rules run on top, catching anything that still slips through and turning it into the correct text deterministically. You add a word once, and it is fixed everywhere you dictate. On iPhone, the same idea lives in the Personal Dictionary, so your custom terms follow you across the apps where you use the keyboard.
And to be clear about what stays private: as of our May 2026 privacy update, the server stores only operational pings. No audio and no transcript content leaves your control, which matters a lot when the words you are adding are client names, case details, or anything confidential.
The dictation app that "learns" your words is not the one with the cleverest hidden algorithm. It is the one that lets you tell it directly, in seconds, and then never forgets.
Tips that compound with a good vocabulary
A custom vocabulary does the heavy lifting, but a few habits make it even more effective:
- Speak the word naturally. Do not over-enunciate or spell it out. The engine matches your normal pronunciation against your vocabulary entry, so say it the way you always say it.
- Capitalize in the entry, not in your speech. You cannot say capitalization, but you can store it. Put the correct casing in the vocabulary so proper nouns come out right.
- Review and prune monthly. Projects end and clients change. A quick monthly pass keeps your dictionary lean and relevant.
- Pair it with cleaner audio. Vocabulary fixes the "which word" problem, but a decent mic and a quiet room fix the "could not hear it at all" problem. For more on this, our dictation tips for better accuracy covers the environmental side in depth.
Frequently asked questions
Does a custom vocabulary slow dictation down?
No. Vocabulary hints and replacement rules are applied as part of the normal transcription process. You will not notice any added delay, and the accuracy improvement on your specific terms is immediate.
How many words can I add?
In practice, far more than you will ever need. Most people get the vast majority of the benefit from their first few dozen entries, because a small set of names and terms causes most of the repeat errors. You can keep adding as new problem words appear.
What if the same word is spelled two different correct ways?
Use a replacement rule to pick the canonical version you prefer, or add the spelling you want as the vocabulary entry. The app will favor the form you taught it.
Will it fix names that sound like normal words?
That is precisely what replacement rules are for. When a name collides with a common word or phrase, a plain vocabulary hint may not be enough, so you map the misheard phrase directly to the correct name. After that it is reliable.
Is this different from autocorrect?
Yes. Autocorrect works on text after it is typed and often fights you on proper nouns. A dictation custom vocabulary works inside the recognition step, so the right word is chosen in the first place rather than being mangled and then half-fixed.
The takeaway
If your dictation app keeps getting the same words wrong, you do not need a smarter app that magically learns on its own. You need an app that lets you teach it, and then you need to spend ten minutes doing exactly that. Build a custom vocabulary from your repeat offenders, add replacement rules for the stubborn ones, and the daily friction that made you doubt dictation in the first place quietly goes away.
Voice Keyboard Pro has a free tier, so you can try Smart Vocabulary on your own hardest words and see how quickly they stop being a problem. Add the five names you mistype most, dictate something that uses them, and watch the difference. Pro is $4.99 a month or $34.99 a year once you are ready for the full feature set.