Guides🎤Voice commandsHands-free with a wake word
Beginner5 min read

Hands-free with a wake word

Start dictating without touching the keyboard. Picking a name, tuning sensitivity, and keeping it from firing at your podcast.

With hands-free mode on, InkSpoke listens for a short name and opens dictation when it hears it. Detection runs entirely on your machine using a local Vosk model — audio is never uploaded for wake-word detection, including on the free tier.

SettingsHands-Free
Hands-free is off by default, and enabling it triggers a one-time ~42 MB model download with an explicit consent prompt. The wake word itself is free — you do not need Pro for it.

Choosing a name

You pick from ten curated names rather than typing any phrase. They were chosen because the offline model recognises them reliably:

Annika · Athena · Cassandra · Miranda · Rebecca · Samantha · Serena · Stella · Victoria · Zeno

The default is Annika. You can optionally add a greeting prefix — hey or hi — so the trigger becomes "hey Annika" rather than a bare name. A prefix noticeably reduces accidental triggers in conversation.

Picking a name derives the other two phrases automatically:

PhraseDoes what
{name} (with optional hey/hi)Opens dictation
thank you {name}Finishes and injects the text
cancel {name}Throws the recording away
A custom name is possible on Pro, but it must be a word the offline model already knows — the settings page shows ✓ Available or ✗ Not found as you type, and refuses to save an unknown word. Downgrading from Pro reverts you to Annika.

Sensitivity

Four levels — Low, Normal, High, and Very High (the default). The setting does one specific thing: High and Very High act on partial results from the recogniser, so they respond sooner but misfire more. Low and Normal wait for the recogniser to finalise a phrase.

It does not slide a confidence threshold. If you are getting false triggers, drop to Normal or add a hey prefix — those are the two levers that actually help.

Keeping it quiet

Several defences are on by default and need no configuration:

  • Speech detection gates the recogniser, so silence and room noise cost almost nothing
  • A steady-noise rejector ignores fans and air conditioning
  • Playback from your own speakers auto-pauses listening, so a podcast saying a similar name will not trigger it
  • After a wake match, a second verification pass re-checks the audio; a failed check shows a "Didn't catch that" notice rather than opening dictation

Turning itself off

Listening stops automatically when your machine sleeps or the screen locks. You can also set an idle timeout (10 minutes if enabled) and a schedule, so it is not live overnight.

Next in this track

Keep going.