GetToasted/rex-wake-words

1# REX Wake-Word Models2 3Custom wake-word models for the [REX voice assistant](https://github.com/David-Antolick/rex_voice_assistant).4 5## Models6 7- **hey_rex.onnx** - openWakeWord model for the phrase "hey rex".8- **hey_rex.tflite** - same model, TFLite runtime version.9 10Trained with the openWakeWord automatic training pipeline:11- 2000 Piper TTS synthetic positives + 329 real recordings from 3 speakers12- Background data: FMA music + ACAV100M precomputed features (~2000 hrs)13- 10,000 training steps on an RTX 3070 Ti14 15## Usage16 17REX 0.3.2+ downloads this model automatically when wake-word gating is enabled.18 19For direct use with openWakeWord:20 21```python22from openwakeword.model import Model23oww = Model(wakeword_models=["hey_rex.onnx"], inference_framework="onnx")24score = oww.predict(audio_int16_chunk_1280_samples)["hey_rex"]25```26 27Audio format: 16 kHz mono int16, fed in chunks of 1280 samples (80 ms).28Default detection threshold: 0.5 (raise to ~0.7 to reduce false fires).29 30## License31 32MIT, matching the REX project. Voice samples used in training were collected33with explicit consent from contributors for inclusion in this model and its34distribution (including any commercial distribution, though none is intended).35