Allow Openwakeword threshold to be in config.json

[cknic]

Right now threshold is hardcoded to > 0.5 in audio_engine.py. (Line 165ish). Can we make this be configurable in config.json. My house is loud since i have a bunch of kids and i'm getting a bunch of false positives on "Alexa" (i know it might just be a poor wake word model) but so i don't have to hack the .py file i'd love if this is configurable. I know the old way of running the LVA command allowed the --wake-word-threshold option.

[imonlinux]

Hey @cknic, This is a great idea. The change is pretty straight forward to implement. Please share your testing results.

edit fix typo in branch name

# Testing: Configurable OpenWakeWord Threshold (branch: `configurable_oww_thresholds`)

This branch adds **configurable OpenWakeWord sensitivity** (activation threshold) to LVA.

The OpenWakeWord trigger check used to be hardcoded to `p > 0.5`. It’s now configurable via:
- a **global** setting in `config.json`, and/or
- an optional **per-model** override in the wakeword model’s `*.json` file.

1) Clone the branch

git clone --branch configurable_oww_thresholds --single-branch https://github.com/imonlinux/linux-voice-assistant

---

2) Global threshold via config.json

In your config.json, add this under the wake_word section:

{
  "wake_word": {
    "openwakeword_threshold": 0.5
  }
}

Notes:

• Allowed range: 0.0 to 1.0 (inclusive).
• If omitted, LVA falls back to the default 0.5.

---

3) Per-model threshold via wakeword model JSON

For OpenWakeWord models, you can override the threshold per model by adding "threshold" to the model’s JSON.

Example file:
wakewords/openWakeWord/ok_nabu_v0.1.json

{
  "type": "openWakeWord",
  "wake_word": "Okay Nabu",
  "model": "ok_nabu_v0.1.tflite",
  "threshold": 0.62
}

Supported per-model keys:

• "threshold" (preferred)
• "openwakeword_threshold" (alias)

This applies only to that specific wakeword model.

---

4) Precedence (which value wins?)

Highest to lowest priority:

1. Per-model wakeword JSON (wakewords/**/<model>.json)
2. Global config.json (wake_word.openwakeword_threshold)
3. Built-in default (0.5)

Optional (if you use it): the CLI flag --wake-word-threshold overrides config.json but does not override a per-model threshold.

---

5) Allowed threshold values (OpenWakeWord)

• Range: 0.0 → 1.0
• Lower = more sensitive (more false positives likely)
• Higher = less sensitive (more missed triggers likely)

Start at 0.5 and adjust in small steps (e.g. 0.05).

---

6) Do I need to restart LVA after changes?

Yes. Restart required for changes to take effect.

• Changing config.json requires restart (config is loaded on startup).
• Changing a wakeword model JSON requires restart (wakeword configs are read and models are instantiated on startup).
• CLI flags only apply at process start.

If you run LVA as a systemd user service:

systemctl --user restart linux-voice-assistant.service

(Use your actual unit name if you changed it.)

---

[imonlinux]

I just merged the sendspin and configurable_oww_thresholds branches into main.

Feel free to open issue if you run into anything.