There is a particular kind of question you do not ask Gemini Notebook. It is the long one — the one where you name three sources, two constraints and the shape of the answer you want. You know it would be a better question than the four-word version. You type the four-word version anyway, because the long one is ninety seconds of typing and you are already mid-thought.
Voice dictation removes that trade. Double-tap Space and hold it, say the whole question out loud, release — and the text is in the chat box, waiting.
How it works
When a notebook is open, a small mic pill sits beside the chat box. You can click it, but the faster path is the hotkey: double-tap Space and hold. Speak while you hold; the words appear in the box as you go, greyed while they are still provisional and settling into place as the recogniser commits to them. Release the key and the dictation stops.
The double-tap matters. A single held modifier would collide with typing — hold Space on its own and you get a run of blank characters — and Ctrl/Alt combinations collide with Vietnamese input methods and with Windows itself. Double-tap-and-hold is the one gesture that never happens by accident while you are writing.
Where the audio goes
This is the part worth being precise about. Dictation uses the browser's Web Speech API: Chrome captures the microphone and sends the audio to Google's speech-recognition service, which returns text. That is the same route any dictation button on any web page takes. Notebook OS sits at the end of it and receives only the text, which it places in the chat box.
Notebook OS never receives the audio, never stores it and never forwards it anywhere. Nothing is uploaded to your notebook until you press send — and what gets sent is text you have read.
Term Drill runs the other direction, and is worth not confusing with this: there the browser reads your terms aloud and pauses for your answer. It does not listen. And Voice Brain Dump, in the side panel, is different again — that one genuinely records an audio file and uploads it to your notebook as a source, the way you would drag in a recording.
What changes in practice
- You ask the long question. Naming the sources, the constraints and the shape of the answer costs fifteen seconds of speech instead of ninety seconds of typing, so you stop economising on the prompt.
- You mix languages without switching keyboards. Pick Tiếng Việt and the English technical terms in the middle of your sentence come through as English; you are not fighting an input method halfway through a thought.
- You keep reading. Dictating a follow-up does not take your eyes off the source you are reading, which is the whole reason you had the follow-up.
The limits, plainly
It is desktop only — Chrome extensions do not run on mobile browsers. Recognition quality is Chrome's, not ours: quiet rooms and ordinary sentences do well, and rare proper nouns still need a glance before you send. The free plan includes 20 dictations a month; Pro removes the limit. The language dropdown offers nine options and remembers your choice.
None of this replaces typing. It removes the moment where a good question turns into a short one because the good version was too much work.