Docs › Writing Studio › Voice input: dictate into the AI chat, punctuation included

Voice input: dictate into the AI chat, punctuation included

Last updated September 10, 2026 · 9 min read

In one line: talk to the AI instead of typing at it — how punctuation gets in, and which mode keeps the audio on your own machine.

The AI Chat panel docked on the right, mid-dictation. The mic button in the input row has gone red with a pulsing dot beside it, and directly above the textarea the grey ghost line reads "Listening… say "comma" or "period" to add punctuation". Once words actually arrive, that same line shows the interim text instead — it sits outside the box on purpose, so nothing you have already typed gets overwritten. The sentence in the box below was dictated, full stop and all.

Scope first: voice input lives in the AI chat input box, not in the manuscript editor. It is for talking to the AI — "chapter three feels rushed, take a look". You cannot dictate prose with it: there is no microphone button in a `.md`, a `.script`, notes or comments.

Starting to talk

Open the AI chat panel. At the right of the input row there is a microphone. Click to start, click again to stop. There is no keyboard shortcut.

  • On a computer: the AI chat panel on the right.
  • On a phone: Chat in the bottom nav, and the sheet that opens.

The first press brings up the browser's microphone permission prompt. Choose Allow.

While you talk, the tentative words appear as a grey line above the input box, not inside it. That is deliberate: if interim text were written into the box, the next interim would overwrite any edit you made by hand mid-sentence. Once a stretch is settled, it is appended to the end of whatever is already in the box — the end, note, not at your caret.

It does not shut off because you paused to think. The browser cuts the session about once a minute; Slima quietly reconnects, so you can keep going.

Punctuation: say it and it appears

Say "comma" and you get ",". This layer is Slima's own, and it works in every case.

Say You get
comma ,
period / full stop .
question mark ?
exclamation mark / exclamation point !
colon :
semicolon ;
ellipsis …
new line / newline a line break
new paragraph a blank line between paragraphs

There is no "dash" command — the word turns up in ordinary sentences too often to be safe. And if the engine also dropped a mark of its own next to the one you spoke, the spoken one wins.

Two paths: the default, and after the offline model

Once you press the mic, recognition can take one of two quite different paths, and their punctuation behaviour is not the same.

The default

  • Uses the browser's own speech recognition service
  • Audio leaves your device to be recognised on the browser's side
  • Needs a network connection
  • The engine adds no punctuation of its own — you speak every mark
  • Desktop and Android

With the offline speech model

  • Recognition runs entirely on your own machine
  • The audio never leaves the device
  • The engine infers punctuation from where you pause
  • Spoken marks still work
  • Desktop only (Mac / Windows / Linux)

How the model gets installed: you do not go looking for a setting. On load, Slima asks the browser whether a speech pack for this language is already there. If not, the download starts the moment you first press the microphone — once, ever. That session still uses the default path, and a line appears below the input: Downloading the offline speech model (one-time). Until it finishes there is no automatic punctuation… The next time you press the mic, it is on the local model.

There is no settings page for the model, no progress bar, no "install now" button, and no way to delete it from Slima — the pack belongs to the browser. Your only cue is the listening line: say "comma" or "period" to add punctuation means the default path, automatic punctuation is on means the local model.

How well the automatic punctuation does varies a lot by language; for Chinese it currently amounts to commas. Arabic has no offline speech pack, so Arabic dictation always takes the default path.

An offline model is not the same as offline dictation. The microphone lives in the AI chat input, and that whole row is disabled whenever the AI chat itself is unavailable — no network, a reply still streaming, or credits exhausted all grey the mic out too. What the local model buys you is privacy and better punctuation, not working without a connection.

A dictation language separate from the interface

Set it under Account → Preferences → Voice input → Voice input language.

Voice input language

  • Follow the interface language (default)
  • 中文（台灣）
  • 中文（简体）
  • English (US)
  • Español (España)
  • العربية (السعودية)
  • 한국어
  • 日本語

These eight options look the same in every interface language: the first is the default, and the other seven are each written in their own language.

There is no Simplified Chinese interface, but you can dictate in Simplified — for the case where you work in a Traditional interface and dictate Simplified text. Pick it and the spoken-punctuation list switches to the Simplified words too, because the vocabulary differs.

The punctuation list follows the dictation language, never the interface language.

This setting is remembered on this device only and does not follow your account. Pick it again on another computer.

Where it does not work

Environment What happens
Desktop Chrome / Edge and other Chromium browsers Fully supported, and this is where the offline model lives
The Slima desktop app The microphone button does not appear. The desktop shell cannot reach a recognition service, and there is no known fix
Firefox No button — the browser keeps this API switched off by default
iOS PWA added to the Home Screen No button. There, the mic lights up but results never arrive, so the entry is removed on purpose — use a Safari tab instead
Android Chrome Works, but there is no offline speech pack

When Slima cannot detect working recognition it declines to draw the button at all, rather than leave an entry point that does nothing when pressed. So "I have no microphone button" is usually not a fault — it is that this environment genuinely cannot do it.

When permission is refused

Errors appear below the input box — nothing fails silently — and the guidance splits three ways:

  • Not granted yet: press the microphone again and choose Allow in the browser prompt; if no prompt appears, reload the page.
  • Blocked: the message names the place to click (on Chromium, the site info icon at the left of the address bar → Microphone → Allow).
  • Allowed but still failing: reload the page; if that does not help, check your operating system has given the browser microphone access.

You may also see No microphone found (unplugged, or another app has it) and Speech recognition could not connect. Once permission flips to allowed, the message clears itself — you do not have to dismiss it.

Related

Try it in Slima

Open the app and do this with your own book. Free to start, no credit card.

Open Slima
Was this helpful?