ovos-ww-plugin-wakeforge

An OpenVoiceOS wake-word plugin. It runs wake-word models made with wakeforge. The models download from Hugging Face on first use.

Install

pip install --pre ovos-ww-plugin-wakeforge

Try it in your browser

Everything runs locally on your machine.

WakeHuBERT wake words: every ready model, on your microphone or an uploaded file.
WakePhoneHuBERT: spot a keyword you type, with no trained model.

Try it on your laptop

The demo listens to your microphone. It runs a ready model next to openWakeWord, microWakeWord and Precise, so you can compare them. Use --help for the options.

pip install --pre "ovos-ww-plugin-wakeforge[demo]"
ovos-wakeforge-demo --wakeword hey_mycroft

The live demo: one probability bar per engine

Use a ready model

Add this to mycroft.conf:

{
  "listener": {
    "wake_word": "hey_mycroft"
  },
  "hotwords": {
    "hey_mycroft": {
      "module": "ovos-ww-plugin-wakeforge",
      "model": "wakehubert_hey_mycroft",
      "listen": true
    }
  }
}

Each ready model is named wakehubert_<word>. The model card lists all words, thresholds, test results and training data: https://huggingface.co/OpenVoiceOS/wakehubert-wakewords.

Threshold

Calibrated models make threshold a sensitivity setting. A higher value gives fewer false activations and fewer detections. A lower value gives more of both. Each model has its own default threshold. Set "threshold" in the hotword config to change it.

WakePhoneHuBERT

WakePhoneHuBERT is a second featurizer, named wakephonehubert-int8. It adds a voice-activity head and an IPA phoneme head to the WakeHuBERT trunk. Models trained on it are named wakephonehubert_<word>. They name their featurizer in their own metadata, so model is the only key you need. The plugin downloads the featurizer on first use.

Use your own model

Set model to the classifier ONNX that wakeforge exported. Set featurizer to its feature-extractor ONNX. Both can be local paths or URLs. A model trained on a pretrained featurizer names it in its metadata, so you can leave featurizer out. Train models with wakeforge.

{
  "listener": {
    "wake_word": "hey_jarvis"
  },
  "hotwords": {
    "hey_jarvis": {
      "module": "ovos-ww-plugin-wakeforge",
      "listen": true,
      "featurizer": "~/hey_jarvis/featurizer.onnx",
      "model": "~/hey_jarvis/model.onnx"
    }
  }
}

Config keys

key	default	meaning
`model`	required	classifier ONNX (path or URL), or a ready model name
`featurizer`	none	feature-extractor ONNX (path or URL), or a pretrained name such as `wakehubert`; optional when the model names one
`threshold`	`0.5`	detection threshold; a ready model or calibrated head uses its own default
`debounce_sec`	`1.0`	minimum seconds between detections (`2.0` for a calibrated head)
`agc`	`false`	level each audio window before scoring; helps when the microphone gain is far too high or low
`models_revision`	pinned	Hugging Face revision of the ready-model repository
`featurizer_revision`	pinned	Hugging Face revision of a pretrained featurizer
`onnx_threads`	`1`	threads per ONNX session

The source documents the other keys (smoothing, patience, window_size, ema_alpha, block_ms, streaming, gru_window, isolated_windows, hidden_dim, vad) in ovos_ww_plugin_wakeforge/__init__.py. URL models are cached under ${XDG_DATA_HOME}/wakeforge/.

Credits

Developed by TigreGótico for OpenVoiceOS.

This project was funded through the NGI0 Commons Fund, a fund established by NLnet with financial support from the European Commission's Next Generation Internet programme, under the aegis of DG Communications Networks, Content and Technology under grant agreement No 101135429.

License

Apache-2.0

Metadata

Release files for ovos-ww-plugin-wakeforge 0.10.0a3

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for ovos-ww-plugin-wakeforge 0.10.0a3
File	Size	Uploaded
ovos_ww_plugin_wakeforge-0.10.0a3.tar.gz	40.7 kB	Oct 4, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for ovos-ww-plugin-wakeforge 0.10.0a3
File	Interpreter	ABI	Platform	Reset
ovos_ww_plugin_wakeforge-0.10.0a3-py3-none-any.whl 29.6 kB Oct 4, 2026	Python 3	none	any	Details

Total release size: 70.3 kB

Release files / ovos_ww_plugin_wakeforge-0.10.0a3.tar.gz

Download URL	ovos_ww_plugin_wakeforge-0.10.0a3.tar.gz
Size	40.7 kB
Tags	Source
SHA-256 checksum How to use checksums	`8509ac52a3ba337065cdf009f298c9ee10d227a647e9196cf90c44c63917c0e8`
BLAKE2b-256 checksum How to use checksums	`9d036365a95edc6fc6d1a653336ce37889d177b1144804526f77df2df3e59d14`
Upload date	Oct 4, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Release files / ovos_ww_plugin_wakeforge-0.10.0a3-py3-none-any.whl

Download URL	ovos_ww_plugin_wakeforge-0.10.0a3-py3-none-any.whl
Size	29.6 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`d614f7f751826957e9087e9db572654b7c62c37903a83de066d5239e1ec47e92`
BLAKE2b-256 checksum How to use checksums	`c6372857ff3875f71478f63612cd629422157562cf476ad4ce585ba680862fd8`
Upload date	Oct 4, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`