How to get fewer errors when you dictate
Most dictation mistakes have boring causes: a laptop microphone half a metre away, a fan humming next to it, a sentence that trails off while you think. Fix those and almost any tool gets noticeably better. This guide starts with a short test that shows which kind of error you make, then goes through the changes in roughly the order they pay off. You will not find accuracy percentages here, because they depend on your voice, your room and your words more than on any brand.
Start with a five-minute diagnosis
Before changing anything, find out what goes wrong. Open a blank note and dictate this paragraph at your normal pace, in one go:
“On Tuesday I spoke with Marta Kowalski about the March budget. She wants the revised figures by Friday, and she asked whether we can move the review to the fourteenth. I said yes, but only if the design team sends their notes first.”
Now mark every error and sort them into four piles. Missing or cut-off words usually point to the microphone, the input level or starting to talk too early. Similar-sounding wrong words (“revised” becoming “revived”) point to pace, mumbling or background noise. Names wrong (“Kowalski”, “Marta”) is a vocabulary problem, and no microphone fixes it. Punctuation wrong is a habit and settings problem. Whichever pile is biggest is where to start; the sections below are ordered to match.
Microphone: distance beats price
The single biggest change for most people is moving the microphone closer to the mouth. A built-in laptop microphone sits near the keyboard and the fan, so it hears your typing, the cooling system and the room as loudly as it hears you. A basic wired headset with a boom microphone a few centimetres from the corner of your mouth puts your voice far above everything else.
Place a boom slightly to the side of your mouth, not directly in front of it, so the puff of air on “p” and “b” does not hit it. For a desk microphone, keep it at a fixed distance, about a hand's width or two, and do not lean back and forth while you talk. Many Bluetooth headsets switch to a lower-quality audio mode when their microphone is active; if a wireless headset gives you more errors than the built-in mic, try a wired one before blaming the software.
- Mac: choose the device in System Settings > Sound > Input and drag the input volume slider. Dictation also has its own Microphone source menu in Keyboard settings; set it to your headset by name if Automatic picks the wrong device.
- Windows: go to Settings > System > Sound, pick the microphone under Input, open its Properties with the arrow, click Start test, speak, then Stop test and Play to hear yourself. Adjust Input volume if the recording is faint or crackles.
Listen to that test recording honestly. If it sounds muffled, distant or full of hiss to you, the recogniser hears the same thing.
Pace and phrase length
Speech models use context: the words around a word help decide what it was. That is why speaking one word at a time, slowly and carefully, often makes results worse. Speak in complete phrases at a steady conversational pace, roughly how you would read a sentence aloud to a colleague.
The trick most people discover late is to think first, then speak. Compose the sentence in your head, then say it in one breath. Long pauses mid-sentence with “um” and restarts give the recogniser broken fragments. If you lose the thread, stop, let go of the key or pause, and start the sentence again cleanly.
Keep sentences moderate. Two clauses is comfortable; a seven-clause sentence dictated in one breath tends to collect small errors near the end, where your voice drops. Ending sentences with energy instead of trailing off helps the last words come through.
Windows voice typing has a Wait time before acting setting; Microsoft says it helps with more accurate recognition whether you speak slowly or quickly. If voice typing cuts you off while you are still forming a sentence, try a longer wait. On a Mac, remember that Dictation stops on its own after 30 seconds without speech.
Names, jargon and vocabulary habits
Recognisers are good at common words and weaker at rare ones: surnames, product names, abbreviations, field-specific terms. A few habits reduce the damage.
- Give rare words context. “Call Dr. Nowak about the scan” is easier than a bare “Nowak” at the start of a sentence.
- Say acronyms as they are normally spoken. If people say “S Q L” letter by letter, do that; if they say “sequel”, expect the tool to write the word, not the letters.
- Keep a short correction list. Write down the five words that always fail. Often there is an alternative phrasing that avoids them, or you can type that one word.
- Do not repeat a failing word ten times. If a word fails twice, type it. Repeating it louder does not teach most tools anything.
- Prefer full words over shorthand you would never write. Dictate “for example”, not “e g”, unless you want the abbreviation.
Punctuation is its own habit. With automatic punctuation on (Mac and Windows both offer a setting), let the tool handle commas and say only the marks it misses. With it off, say every mark. Mixing the two styles gives doubled or missing punctuation.
The room you dictate in
Hard, bare rooms echo, and echo blurs consonants. A room with curtains, a rug, a sofa or a bookshelf sounds drier and works better. Facing a wall at close range is worse than facing into the room.
Steady noise (a fan, an air conditioner, a fridge) is easier to cope with than voices. A television or a colleague talking nearby is the worst case, because the recogniser cannot easily tell which voice is yours. Windows voice typing uses echo cancellation so it listens mainly to your voice, but distance is still your best tool: the closer the microphone, the less the room matters.
Check your own computer too. A laptop working hard spins its fans right next to the built-in microphone. Closing a heavy browser tab or a video call app can quieten things more than you would expect.
What moves results, and how much effort it takes
| Change | Fixes mostly | Effort |
|---|---|---|
| Headset or close microphone instead of laptop mic | Missing words, wrong similar words | Low, one-time purchase |
| Correct input device and input volume | Cut-off or faint words | Two minutes |
| Think, then say the whole phrase | Fragments, restarts, odd word choices | A week of habit |
| Quieter, softer room | Wrong similar words, random insertions | Varies |
| Context around names and jargon | Names and rare terms | Habit |
| One punctuation style, consistently | Doubled or missing marks | Habit |
| Switching to a different dictation tool | Depends on the tool and your languages | An afternoon of testing |
Notice that changing tools comes last. It can make a real difference, especially for mixed languages (see dictating in multiple languages), but a poor microphone hurts every tool equally.
Things that usually do not help
- Shouting. Louder input often distorts; a normal voice close to the microphone is cleaner.
- Robot speech. Separating every word removes the context models rely on.
- Correcting mid-flow. Stopping to fix each error breaks your rhythm. Dictate the paragraph, then edit it in one pass.
- Changing five settings at once. Change one thing, repeat the test paragraph, compare. Otherwise you never learn what worked.
When the tool itself is the limit
If your recordings sound clear and your habits are good but results are still poor, the recogniser may not fit your language, accent or vocabulary. That is the moment to try another one, using the same test paragraph so the comparison is fair.
Voice Monkeys is one option to include in that test. On Apple Silicon Macs it runs the speech model on the Mac's Neural Engine, offline once a one-time model download of about 600 MB is done; on Windows, Linux and Intel Macs it uses a cloud server in Germany. It recognises 25 languages without switching. There is a 15-second online demo on the site (five tries a day) and a 7-day trial from the download page. Run the same paragraph through it and through your current tool, and keep whichever needs fewer fixes for your voice. If you work mostly in a language it does not list, a built-in tool or another app will serve you better. For a wider comparison, see the best dictation apps, and for problems where dictation does not start at all, see the troubleshooting sections in dictation on Mac and dictation on Windows.
Questions and answers
Does a more expensive microphone always give better dictation?
No. Distance matters more than price. An inexpensive wired headset close to your mouth usually beats an expensive desk microphone across the table.
Should I speak slowly for dictation?
Speak at a steady, normal pace in complete phrases. Very slow, word-by-word speech removes the context that helps the recogniser choose the right word.
Why does dictation keep getting one name wrong?
Rare names are the weakest point of most recognisers. Add context around the name, pronounce it the usual way, and if it still fails twice, type it instead of repeating it.
Does background music affect dictation?
Music with vocals is the hardest kind of noise, because the recogniser hears another voice. Instrumental music at low volume matters less, and a close microphone reduces both.
What accuracy can I expect?
That depends on your microphone, room, language and vocabulary far more than on any published figure. Dictate the test paragraph from this guide and count your own errors before and after each change.
Sources
- Apple: Use Dictation on Mac (checked November 6, 2026)
- Apple: Change the sound input settings on Mac (checked November 6, 2026)
- Microsoft: Use voice typing to talk instead of type on your PC (checked November 6, 2026)
- Microsoft: How to set up and test microphones in Windows (checked November 6, 2026)