Ask anyone who tried dictation software a decade ago why they gave up, and a surprising number will describe the same thing. Not accuracy. Punctuation. You would talk for a minute and get back a wall of lowercase words with nothing separating them, and spend longer fixing it than you would have spent typing it.
That problem is genuinely solved now, and the solution is more interesting than it first appears.
The short version: PeekoType punctuates and capitalises as you go, without being asked. Every burst of dictation is finished with a full stop, sentences start with capital letters, and text after a full stop or question mark is capitalised too. You can still speak punctuation when you want a mark in a particular place, but you no longer have to.
Punctuation is a comprehension problem, not a typing problem
Here is what makes this harder than it sounds. Punctuation is not in the audio. Nobody pronounces a comma. When you say "right so the thing is we need to move the meeting", the words carry no marks at all. Where the full stop belongs depends entirely on what the sentence means.
Older speech engines matched sounds to words and stopped there, which is exactly why they produced those unbroken walls of text. Modern models work differently. They were trained on vast quantities of written and spoken language together, so they have learned where punctuation tends to go in real sentences. They are not detecting pauses. They are understanding structure.
That is why the result reads like writing rather than like a transcript of someone thinking out loud.
What PeekoType does without being asked
- Finishes your sentences. If a burst of dictation ends without a full stop, question mark or exclamation mark, one is added. Short utterances are exactly where speech models most often leave the ending off, so this quietly catches a lot.
- Capitalises the first word. Every dictated chunk starts with a capital, so consecutive bursts chain together as proper sentences rather than mid-sentence fragments.
- Capitalises after sentence endings. Text following a full stop, question mark or exclamation mark gets a capital, so a paragraph dictated in one go is internally consistent.
- Leaves your commands alone. If you deliberately said new line, or dictated something that is pure punctuation, it does not helpfully "fix" it into something you did not ask for.
The practical effect is that you can dictate three sentences into an email, one at a time, over the course of a minute, and get three properly formed sentences rather than three fragments you then have to stitch together.
When to take over manually
Automatic punctuation is a good default, not a mind reader. There are moments when you want a specific mark in a specific place, and for those you simply say it out loud.
A comma changes meaning and the model cannot always know your intent. A question mark on a sentence phrased as a statement, which happens constantly in English, needs telling. A line break in the middle of a thought is a formatting decision, not a grammatical one. Semicolons and colons are stylistic and the model will rarely guess them.
All of these are covered by spoken voice commands, which sit alongside the automatic behaviour rather than replacing it. Say nothing and you get sensible defaults. Say "comma" and you get a comma exactly where you wanted one.
The filler word problem, and why it matters here
Punctuation is only half of what makes a transcript readable. The other half is everything you said that you did not mean to say.
Real speech is full of "um", "uh", "you know" and "sort of". They are invisible in conversation and painfully obvious in text. Punctuation cannot help with them, which is why filler word removal exists as a separate feature. Together the two turn a raw stream of speech into something you could send without editing.
Does it work in other languages?
Yes, and this catches people out pleasantly. The model punctuates whatever language it is transcribing, following that language's conventions. Dictate in French and you get French punctuation and capitalisation. The same holds across all 100 languages PeekoType supports.
Getting the best from it
Two small habits make a noticeable difference.
First, dictate in thoughts rather than in paragraphs. Say a sentence or two, pause, carry on. This is not a technical requirement, it is simply how the model sees natural boundaries most clearly, and it happens to match how people speak anyway.
Second, do not fight it. People new to dictation often try to say every mark aloud out of habit from older software, then find themselves with doubled-up punctuation. Try a paragraph saying nothing but the words. Most people are surprised by how little tidying it needs.
If you want the full picture of how dictation fits into daily work, our practical daily guide covers the habits that make it stick, and the user guide documents every option in the app.