ovos-ww-plugin-openwakeword 0.4.8a1

Description

This plugin adds openWakeWord support to OpenVoiceOS (OVOS). openWakeWord is an open-source wake word detection system. It trains on synthetic data and runs on a single core of a Raspberry Pi 3.

Install

On Python 3.11 and older, run this command:

pip install ovos-ww-plugin-openwakeword

On Python 3.12 and newer, a plain install fails to resolve: openwakeword requires tflite-runtime, and tflite-runtime has no wheels for these interpreters. Install openwakeword without its declared dependencies, then let this plugin provide the dependencies the onnx path needs:

pip install --no-deps "openwakeword>=0.5.0,<1"
pip install "ovos-ww-plugin-openwakeword[onnx]"

Set your wake word in mycroft.conf. Do not set the models key if you want the default model ("hey jarvis").

"listener": {
    "wake_word": "hey_jarvis"
},
"hotwords": {
  "hey_jarvis": {
      "module": "ovos-ww-plugin-openwakeword"
  }
}

Configuration

The plugin accepts these extra options:

"listener": {
    "wake_word": "hey_jarvis"
},
"hotwords": {
  "hey_jarvis": {
      "module": "ovos-ww-plugin-openwakeword",
      "models": ["path/to/openwakeword/model/hey_jarvis.tflite"],
      "inference_framework": "tflite",
      "custom_verifier_models": {"hey_jarvis": "path/to/openwakeword/custom/verifier/model.pkl"},
      "threshold": 0.3,
      "custom_verifier_threshold": 0.1
  }
}

• models: paths to one or more openWakeWord models, in .onnx or .tflite format. Any model in the list can activate OVOS.

• inference_framework: the format of the models in models. Use tflite or onnx. The plugin picks tflite when tflite-runtime is installed and onnx when it is not — on Python 3.12+ tflite-runtime is usually not installable, so onnx is the effective default there. openwakeword itself falls back to onnx at runtime when tflite-runtime is absent, even if tflite was requested.

• threshold: the score needed to trigger activation. Higher values need a stronger match. The default, 0.5, works for most cases.

• custom_verifier_model and custom_verifier_threshold: paths and settings for custom verifier models, supported since openWakeWord>=0.3.0. A custom verifier model can improve performance when the included pre-trained models do not fit your deployment.

See the openWakeWord repository for more details.

Related projects

• openWakeWord: the wake word engine this plugin wraps.
• OVOS Plugin Manager: loads and configures this plugin.
• ovos-ww-plugin-precise-lite: another OVOS wake word plugin.

License

Apache-2.0