pymicro-wakeword 2.3.0

A TensorFlow based wake word detection training framework using synthetic sample generation suitable for certain microcontrollers.

Python microWakeWord

Python library for microWakeWord.

Uses a pre-compiled Tensorflow Lite library.

Install

pip3 install pymicro-wakeword

Usage

from pymicro_wakeword import MicroWakeWord, MicroWakeWordFeatures, Model

mww = MicroWakeWord.from_builtin(Model.OKAY_NABU)
mww_features = MicroWakeWordFeatures()

# Audio must be 16-bit mono at 16Khz
while audio := get_10ms_of_audio():
    assert len(audio) == 160 * 2  # 160 samples
    for features in mww_features.process_streaming(audio):
        if mww.process_streaming(features):
            print("Detected!")

Use process_streaming_prob instead to get the wake word probability. If this probability is greater than probability_cutoff, the wake word is detected.

Command-Line

WAVE files

python3 -m pymicro_wakeword --model 'okay_nabu' /path/to/*.wav

Live

arecord -r 16000 -c 1 -f S16_LE -t raw | \
  python3 -m pymicro_wakeword --model 'okay_nabu'