Skip to main content

Enable voice dictation

For any operator who wants to talk to their AI instead of typing: set up a keyboard shortcut that transcribes your voice straight into the chat field, locally.

Prerequisites​

  • A dictation model (Whisper) is downloaded via Download local models.
  • Your machine's microphone works and Apollia is allowed to use it.
  • A free keyboard shortcut that does not conflict with the OS or another application.

Steps​

  1. In the sidebar, click Settings, then the Dictation section. Settings, Dictation section, state of the dictation model shown at the top

  2. Check that a model is named under Active model. It loads on first use. If none is named, download a dictation model from the AI screen, or click Load a model to pick a model file you already have.

  3. Select the dictation Language in the picker. It offers thirteen languages plus Auto-detect; the exact list is in Configuration. Naming the language improves accuracy noticeably, and auto-detection is unreliable on short recordings.

  4. Select your microphone in the Microphone picker. Leaving it on the system default is fine until a virtual device (BlackHole, Soundflower, an aggregate device) becomes that default, in which case Apollia records silence. Naming the device removes the ambiguity.

  5. Click the Global hotkey field. A window prompts you to press the key combination you want (for example Cmd + Shift + Space). HotkeyCapture window with the message "Press your hotkey combination" and the captured combination

  6. In the Trigger mode picker, choose one of the two modes:

    • Toggle (press = start/stop): a first press on the shortcut starts recording, a second one stops it.
    • Push-to-talk (hold): you hold the shortcut down while you speak, and transcription starts when you release it.

    For the rest of this procedure, choose Push-to-talk (hold).

  7. Save. Dictation settings apply to the next dictation; there is nothing to restart.

  8. Before leaving the page, press Test and speak a few words. The bars must follow your voice, and the recognised text appears underneath. Flat bars mean the selected device is delivering nothing.

  9. Open a chat from the sidebar.

  10. Hold your shortcut down. A full-screen dark overlay appears with a bar audio visualiser. Speak naturally. On screen: the full-screen recording overlay, with the bar audio visualiser and the text {hotkey} to stop · Esc to cancel.

  11. Release the shortcut. The transcription is injected into the input field through the clipboard.

    Note: the transcription is inserted by simulating a paste (Ctrl+V / Cmd+V). The input field must be focused to receive the text.

  12. Read it back, fix it if needed, then press Enter to send.

Verification​

A spoken sentence of a few seconds shows up transcribed in the input field, exactly once, without any data leaving your machine.

What happens when nothing was heard​

Silence is not transcribed. When every moment of a recording sits below the silence threshold, Apollia says that nothing audible was captured instead of sending the audio to the model. This matters because a speech model handed silence does not return an empty result: it returns plausible sentences it was never given, and those used to reach the operator looking exactly like real transcriptions.

If it does not work​

  • No transcription: name your microphone in the Microphone picker rather than relying on the system default, then read Voice dictation transcribes nothing.
  • Shortcut ignored: another application may be capturing the same combination; pick a less common one.
  • Rough transcription: check that the selected language matches the one you speak, and consider a larger Whisper model.
  • The text does not show up in the field: make sure the chat input field is focused (click it) before using the shortcut.

Technical reference: Apollia reference - supported engines, Whisper model sizes, audio formats, latency optimisations.