Windows already has dictation. So why something else?
Fair question, and the answer starts with something nobody likes to admit: for plenty of jobs, Windows dictation is enough. It’s free, it’s already installed, and it opens with Win+H. If you dictate a couple of lines of plain English with a connection up, it does the job.
This page is about where it doesn’t reach, using facts from Microsoft’s own documentation, so you can decide whether those limits apply to you.
Six things Win+H doesn’t do
These aren’t bugs: they follow from how it’s built. None of them can be worked around with a setting.
It wants a connection
Microsoft’s requirements page says it in one line: to use voice typing you need to be connected to the internet. Recognition runs on Azure speech services, so your voice leaves the computer.
Here: with the local engine, transcription happens on your own computer and the audio never leaves it. It works on a plane.
There’s no dictionary of yours
There is no user-editable dictionary. Client names, acronyms, trade terms and brand names come out however the model guesses, every time, with no way to teach it the right spelling.
Here: the words you teach it beat everything else — even a famous brand name the model would rather write instead.
You can’t fix it by speaking
There are voice commands, but few and generic: new line, delete that, punctuation spoken aloud. To fix that one wrong word in the middle of a sentence, you go back to the keyboard.
Here: press the correction key, say the word the way it should be spelled, and it remembers it from then on.
It never says it wasn’t sure
If it mishears a name, it writes it and moves on. You find the mistake by proofreading — or you don’t, which is worse, because it leaves in an email.
Here: when a word isn’t certain it tells you (“cTrader, did you mean Citrader?”) and one click settles it.
The good rewriting needs a new PC
Auto-punctuation is on every PC, but the layer that genuinely rewrites speech into clean prose is reserved for Copilot+ machines with a dedicated neural processing unit.
Here: the cleanup — hesitations dropped, second thoughts understood — works on any Windows 10 or 11.
One language at a time
You pick a language and it stays picked. Drop three foreign technical terms into a sentence — normal if you work in tech — and they get transcribed by ear.
Here: over a hundred languages, and mixing inside one sentence comes out spelled right instead of mangled.
The Win+H limits above are taken from Microsoft’s support page on voice typing and are accurate as of 27 July 2026. Microsoft updates Windows often: if something here is no longer true, tell us and we’ll fix it.
When Windows is the right choice
It wouldn’t help us to pretend otherwise: there are plenty of people for whom paying for this makes no sense.
Win+H is enough
If you dictate rarely, and almost always in one language.
If your writing has no proper nouns, acronyms or technical terms that must be spelled a particular way.
If you’re always connected and your voice passing through a cloud service doesn’t bother you.
If you have a Copilot+ PC, where speech rewriting is already included.
You need more
If you write client, library or product names every day and they have to be right first time.
If you’ve had a corrector “fix” a name that was already correct.
If the audio of what you say must not leave your computer.
If two languages get mixed inside your sentences.
A corrector may change how a word is spelled.
Not which word you said.
It’s the rule that separates us from any generic dictation.
The fastest way to decide is to try it
Seven days with everything switched on. If after a week Win+H still feels like enough, you have your answer and you haven’t paid a thing.
Seven-day trial · Nothing charged before it ends · Cancel in one click