Why Punctuation Is the Hard Part of Dictation

Modern speech to text gets the words right almost every time. Ask anyone who has tried Mac dictation what actually annoys them, though, and it is rarely a wrong word. It is the wall of text with no commas, no periods and no paragraphs. Punctuation, not vocabulary, is the real hard part of dictation.

Words are easy, punctuation is not

A speech-to-text model is trained to map audio to the words it hears. Local engines like Whisper and Parakeet are genuinely good at this now, even offline, and they handle a surprising range of voices. The trouble is that punctuation is not something you hear. A period is not a sound. A comma is not a sound. A new paragraph is definitely not a sound. So when your voice to text engine turns audio into words, it has almost nothing to anchor a comma to.

This is different from the accuracy problem people expect. You can read more about how models cope with different voices in our piece on how speech to text handles accents, but even a perfect word-for-word transcript still arrives with zero structure. The words are all correct, and the text is still unreadable.

Why pauses are a bad clue

The obvious workaround is to insert punctuation where the speaker pauses. Basic dictation engines do exactly that, and it is why raw Mac dictation feels random. People do not pause at grammatical boundaries. We pause to think, to breathe, to search for a word. We rush straight through the actual end of a sentence and then stop dead in the middle of the next one.

So pause-based punctuation drops commas inside phrases that should stay together and misses the period where a thought clearly ends. The result is the classic run-on: one endless sentence that technically contains every word you said and none of the shape you meant.

Saying the punctuation out loud is not the answer

Classic dictation, including Apple's built-in tool, lets you speak the marks: say "comma," say "period," say "new paragraph." It works, and some heavy users swear by it. But it asks you to be a typesetter while you are also trying to think. You have to hold the full punctuation of a sentence in your head before you say it, which is the opposite of talking naturally. Most people speak around three to four times faster than they type, and the whole point of dictation is to keep that speed. Narrating every comma throws it away.

This is also where voice control for power users splits from everyday dictation. Command-driven tools such as Talon Voice give you precise spoken control, which is powerful for hands-free workflows and one reason people learn to code by voice on a Mac. For writing prose, though, most of us just want to talk and get clean sentences back.

The fix: let AI read meaning, not pauses

The reliable way to punctuate speech is to stop guessing from audio timing and instead look at the words themselves. A language model can read a finished transcript and place punctuation based on grammar and meaning, the same way you would if you were editing it. It knows where a clause ends because it understands the clause, not because you happened to breathe there.

This is what BlaBlaType's on-device AI cleanup does. After the local model transcribes your voice, Apple Intelligence rewrites the raw text: it adds the punctuation, removes filler like "um" and "you know," fixes obvious grammar and keeps your tone. If you want the technical version of this, we broke it down in how AI rewriting works without the cloud. The important part is that both steps run on your Mac, so your audio and text never leave the device.

Same words, but now it is readable. Notice that the cleanup did more than sprinkle commas: it split one run-on into three sentences, capitalized the name, and turned a statement into the question it always was. That is the difference between guessing at pauses and understanding the sentence.

Why this matters beyond tidy text

Clean punctuation is not just cosmetic. If dictation forces you to stop and fix commas after every paragraph, you lose the speed advantage that made you talk instead of type in the first place. For people who dictate to protect their hands from typing strain, that matters even more: the NHS guidance on repetitive strain injury points to reducing repetitive keyboard work, and dictation only helps if you are not immediately back on the keyboard cleaning up.

It also matters for privacy. The easiest way to get great punctuation is to send your transcript to a cloud model, but then your words leave your Mac. Doing the cleanup on-device keeps the quality without the trade-off. If that is a priority for you, see whether Mac dictation is actually private, and check the plans to see what runs locally.