Dictation
Hold right Option and talk. Your words land where the caret is.
Dictation is off until you switch it on in Settings → Dictation. Then hold the right ⌥ key, talk, and let go: the words are put where your caret is.
The keys
- Hold to Talk records while the key is down. Start and Stop is a toggle: one press starts, the next stops.
- Click the panel to stop, or its mode name to switch modes.
- When Finished is Paste (the default) or Copy. With Copy, hold ⇧ as you finish to paste this time.
- Recordings stop by themselves after 10 minutes. Anything under 0.3 s is dropped.
- The words are also left on the clipboard every time, pasted or not.
The panel
While you talk, a small panel shows you're recording. At the top of the screen it grows out of the notch and shows your words below it as you speak; anywhere else it's a glass capsule with your words in one running line.
Drag it anywhere. It settles in the nearest of six places (top or bottom; left, middle or right) and appears there from then on. Drop it at the top middle to put it back in the notch. If another app already uses the notch, the panel sits just below it.
To hide your words while you talk, for example while sharing your screen, turn off Show Words While Talking.
Local or cloud
Speech is written down on your Mac by a model you download, or by a server you choose.
Parakeet v3 Neural Engine is the one to start with. It runs on your Mac's Neural Engine and covers 25 European languages. On a MacBook Air it wrote down a 56-second recording in about a second, where the same model on the CPU took 5.7 s. It's a 570 MB download.
Others are there for other needs: Parakeet v2 and 110M (English), Cohere Transcribe, Qwen3-ASR, Canary, Moonshine, SenseVoice, and Whisper in five sizes for the languages the others don't cover.
A local model hears you as you speak. At every pause it writes down what you've said so far, so when you let go the words are ready almost at once.
Your Mac's sound
Other Audio decides what happens to whatever's playing while you talk. It starts on Leave, which leaves it alone.
- Muffle turns it down (macOS 14.2 and later; it asks to record system audio, and plays the sound straight back).
- Mute silences it.
- Pause presses play/pause, and only when another app is playing sound.
It's put back when you stop. Pick a microphone under Sound → Microphone, or leave it on System to follow your Mac's Sound settings. If the one you picked is unplugged, Tabussi uses the system microphone until it's back.
Remove Silence and Even Out Volume (both on) tidy the recording before the model hears it.
Listen for Names on Screen (on) reads names in the window you're dictating into, so they're spelled right. They're read on your Mac and never kept.
Memory and battery
Settings → Dictation → Models → Advanced. The speech model unloads after 5 minutes unused (Unload After: each use, 5 min, 30 min or never), and straight after each use on battery (Unload on Battery, on).
History
Settings → History lists what you dictated, by day, in an encrypted file on your Mac: search it, copy a dictation, see the original before a rewrite, and where the wait went. Recordings are never saved. Delete one or all, or switch History off from its banner.
Next
- Modes: a language, an AI rewrite and a shortcut, per kind of writing.
- Vocabulary: names and terms spelled your way.