Voice typing
Speak into the AI chat or the command center, with a speech model that runs on your computer.
The mic left of Send in the AI chat's prompt writes down what you say at the caret. The speech model runs on your own computer, so the recording never leaves it, it works offline, and it costs nothing.
Setting it up
The first press on the mic downloads the speech model: Parakeet, about 670 MB, which understands 25 European languages and adds punctuation and capitals. Progress shows around the mic. Every profile shares the one download.
Right-click the mic to cancel a download, remove the model, or open your system's microphone settings.
Speaking
| How | Does |
|---|---|
| Press the mic, speak, press ✓ | Writes it down. ✕ drops it. |
| Hold the mic, speak, let go | The same, in one press |
| ⌘Ctrl⇧ShiftD with the prompt focused | Starts or stops, like the mic |
| Hold Space in the prompt | Records until you let go |
| Esc | Drops the recording |
A waveform shows while it's listening. A recording can be up to 2 minutes long, and the text shows well under a second after you stop.
In the command center
Once the model is downloaded, hold Space in the command center's search box to speak a search, an address, or a chat message. The space you typed is taken back, and letting go writes down what you said. Holding Space never starts the download, so get the model from the chat's mic first.
Which microphone
Settings → Audio, mic, and SFX → Microphone preference lists every microphone the browser has seen. Voice typing records from the first one on the list that's plugged in, else the system's default, so a Bluetooth headset takes over whenever it's connected. Drag to reorder, or press × to remove one.