wakeword-trainer
Copyright 2026 wakeword-trainer contributors

This product includes software and models developed by third parties.

--------------------------------------------------------------------------
openWakeWord
--------------------------------------------------------------------------
Portions of the training recipe, model architecture, feature-extraction
pipeline, and the bundled ONNX feature models are derived from or compatible
with the openWakeWord project:

    https://github.com/dscripka/openWakeWord
    Copyright 2022 David Scripka
    Licensed under the Apache License, Version 2.0

The bundled files `melspectrogram.onnx` and `embedding_model.onnx` originate
from the openWakeWord project and are redistributed unmodified under the
terms of the Apache License, Version 2.0.

--------------------------------------------------------------------------
Google speech_embedding
--------------------------------------------------------------------------
The `embedding_model.onnx` implements the Google `speech_embedding` model
(https://tfhub.dev/google/speech_embedding/1), and the melspectrogram model
approximates its front-end. Both are distributed by Google under the
Apache License, Version 2.0.

--------------------------------------------------------------------------
augmentations
--------------------------------------------------------------------------
Optional audio augmentation behaviour is inspired by the augmentation
recipes used by openWakeWord, audiomentations and torch-audiomentations.
No source code from those projects is vendored in this repository.

--------------------------------------------------------------------------
ACAV100M features
--------------------------------------------------------------------------
Pre-computed negative features are downloaded on demand from the
openWakeWord feature dataset hosted by David Scripka on Hugging Face:

    https://huggingface.co/datasets/davidscripka/openwakeword_features

These files are not included in this repository or in published wheels.

--------------------------------------------------------------------------
Piper, Coqui XTTS, Silero TTS
--------------------------------------------------------------------------
TTS support is optional and is not bundled. Any TTS engine, voice model,
or custom backend installed by the user remains subject to its own license
terms. Users are responsible for complying with those terms, including any
non-commercial or attribution requirements.
