voice-dictation-setup

name	voice-dictation-setup
description	Opinionated end-to-end setup for clean voice dictation input — confirm Easy Effects is running, ensure a noise-reduction + autogain preset is active, run a level test, and report whether the chain is dictation-ready.
needs_user_vars	[{"name":"EE_LIBRARY_PATH","type":"path"}]

voice-dictation-setup

A one-shot guided flow for "make my mic dictation-ready." Composes other skills in this plugin; doesn't reimplement them.

Steps

1. Confirm Easy Effects is running

if command -v easyeffects >/dev/null 2>&1; then
  EE_RUN="easyeffects"
elif flatpak info com.github.wwmm.easyeffects >/dev/null 2>&1; then
  EE_RUN="flatpak run com.github.wwmm.easyeffects"
else
  echo "Easy Effects not installed."; exit 1
fi
pgrep -f easyeffects >/dev/null || $EE_RUN --hide-window &
sleep 1

2. Identify the active mic

pactl get-default-source

Confirm with the user that this is the mic they want to use for dictation. If wrong, point them at pavucontrol or KDE / GNOME sound settings to switch the default, then resume.

3. Ensure a noise-reduction preset is loaded

Run the list-presets skill (mentally — don't recursively invoke; just call its commands inline) and look for a preset whose name suggests it includes:

• An RNNoise / NoiseTorch / DeepFilter / Speex stage (denoise)
• A high-pass at ~80–100 Hz
• An autogain or compressor stage

Common names: "Speech Processing", "Voice Dictation", "Clean Voice", "NR + AutoGain". If none exists in $EE_DATA/input/:

• Check the user's library ($LIB/input/) — if a candidate is there, call install-preset.
• If nothing suitable exists anywhere, surface the gap to the user and offer to create a basic preset. (Authoring presets is best done in the Easy Effects GUI; the plugin doesn't try to synthesize one.)

Once a candidate is identified, load it via the load-preset skill pattern:

$EE_RUN -l "$preset_name"
$EE_RUN -a input

4. Test levels

Run the test-input-level skill flow with duration_sec=6. Ask the user to read a sentence aloud during the window.

5. Verdict

Tell the user one of:

• Ready — preset loaded, peak in [-12, -6] dB, RMS in [-20, -16] dB, no DC offset issue. Dictation should perform well.
• Loaded but too hot/quiet — preset is fine, but gain is wrong. Suggest the specific adjustment (autogain target, OS mixer level). Offer to re-test.
• No suitable preset — user needs to author one in the GUI. Sketch the chain (HPF → RNNoise → Autogain) and stop.

6. Optional: bind it to the mic

If the user is happy and wants this to load automatically for that mic from now on, offer to call bind-mic-autoload with the chosen preset and the active source's node.name + port.