How to dictate punctuation cleanly
Punctuation is the part of dictation people worry about most. They picture themselves narrating "comma" and "period" like a court stenographer, and decide the whole thing sounds exhausting before they even start. The good news is that modern dictation does not really work that way anymore.
This post walks through what to expect from punctuation today, when it pays to say a mark out loud, and a few habits that make your dictated text come out clean enough to use without much cleanup.
Modern models infer most of it
The old advice — "say every comma" — comes from an earlier era of speech recognition, when the software had no sense of sentence structure and simply transcribed whatever sounds it heard. Today's speech models are different. They are trained on enormous amounts of real language, so they have a strong sense of where sentences end and where clauses break.
In practice this means a current model will, on its own, do most of the following:
- End a sentence with a period when your pitch falls and you pause.
- Start the next word with a capital letter.
- Drop a comma into a natural mid-sentence pause.
- Add a question mark when your phrasing rises like a question.
- Capitalize obvious proper nouns and the word "I".
You do not have to ask for any of that. If you speak in ordinary, complete sentences with the pauses you would naturally use when talking to a person, the punctuation tends to land in roughly the right places. For a closer look at how this works, see automatic punctuation.
So the first rule of dictating punctuation cleanly is almost the opposite of the old advice: stop thinking about punctuation, and think about phrasing instead. Speak the sentence the way you would say it aloud. The model is listening to your rhythm, and your rhythm carries most of the information.
Your pauses are the real instructions
Because the model reads your delivery, the way you pace a sentence becomes a kind of punctuation in itself.
- A short pause in the middle of a thought reads as a comma.
- A full stop in your voice — the pitch drop, the breath — reads as a period.
- Running two ideas together with no pause reads as one long clause, which is often where dictated text gets messy.
That last point is the one worth practicing. When people speak too fast, they merge sentences that should be separate, and the model has no clear seam to cut on. The fix is not to slow every word down. It is to give a genuine, deliberate pause at the end of each sentence — about a beat longer than feels necessary. That single habit cleans up more punctuation problems than any spoken command.
When to say the mark out loud
Inference handles the common cases. It cannot read your mind on the cases that depend on meaning rather than rhythm. For those, saying the mark explicitly is the right move.
Spoken punctuation is genuinely useful for:
- Colons and semicolons. These rarely have a distinct sound in speech, so the model often will not place them. If you want one, say "colon" or "semicolon".
- Parentheses and quotation marks. Say "open paren" and "close paren", or "quote" and "end quote". The model cannot guess where a bracketed aside begins and ends.
- Dashes and ellipses. Say "dash" or "em dash" and "dot dot dot" when you want that specific texture.
- Lists and line breaks. "New line" and "new paragraph" are the most valuable spoken commands of all, because layout has no sound at all. Nothing in your voice tells the model you want a fresh paragraph.
A reasonable mental model: let the model handle sentence-level punctuation — periods, commas, capitals, question marks — and reserve your voice for the structural marks it cannot hear. That keeps dictation feeling natural while still giving you control where it counts.
A clean-dictation routine
Here is a simple way to put this together when you are dictating a paragraph or two.
- Decide the sentence before you speak it. A half-formed sentence comes out as a half-formed sentence. A moment of silent thought is cheaper than a cleanup later.
- Speak the sentence in one even pass, at conversational pace.
- Pause clearly at the end. Let the period land.
- Say "new paragraph" when you change topic.
- Say a punctuation mark out loud only when it is a colon, a bracket, a quote, or a layout break.
- Read the result once at the end, not after every sentence.
That last point matters. Stopping to inspect punctuation after each sentence breaks your train of thought and usually makes the writing worse. Dictate the whole passage, then read it back once. Most of the time the only fixes are a stray comma or a missed paragraph break.
Handling the awkward cases
A few situations trip up almost every dictation tool, and they are worth knowing in advance.
Numbers and units. "Five point two" usually becomes "5.2", but spacing around symbols like percent or currency can be inconsistent. If a document is heavy with figures, expect to touch those up by hand.
Abbreviations with periods. "e.g." and "i.e." are hard to dictate cleanly because the periods inside them look like sentence boundaries. It is often faster to dictate "for example" and "that is", or to type those two abbreviations yourself.
Run-on questions. A long question that trails off without a clear upward inflection may come back with a period instead of a question mark. If the question mark matters, give the end of the sentence a small, honest rise.
None of these are failures of dictation. They are simply the edges where speech and written convention do not line up neatly. Knowing where they are means you can glance at them on review instead of being surprised.
When a mark name is the word you want
One genuinely confusing case is when you want to dictate the word "comma" rather than a comma. Most tools, most of the time, will insert the punctuation. If you are dictating about punctuation itself — or any text where mark names appear as words — expect to fix those by hand. It is a rare enough case that it is not worth changing your habits for. Just know it exists.
The honest summary
Clean dictated punctuation is mostly a speaking skill, not a command vocabulary. Speak in complete sentences, pause honestly at the ends, and let the model do the sentence-level work it is good at. Save your spoken marks for colons, brackets, quotes, and paragraph breaks — the things your voice physically cannot signal.
With Lispr, you hold the right Option key, speak a sentence with its natural rhythm, and release. The Whisper model handles the periods and commas; you stay focused on what you are actually trying to say. Practice the deliberate end-of-sentence pause for a day, and clean punctuation stops being something you think about at all.
Try Lispr
Voice to text in any Mac app — hold a key, talk, let go. Free, no account, ~4 MB.
Download for macOS