KWS¶

Keyword Spotting (KWS) or Keyword Search (alsa called Keyword Detection, Wake Word Detection) is a key technology to implement a hands-free voice assistant. It's used to detect words such as "OK, Google", "Hey, Siri" to start a conversation. Unlike speech recognition, KWS is a always-listening program which should be run locally without and offline.

There are a variety of KWS projects, such as Snowboy, Mycroft Precise, Porcupine.

Snowboy¶

1. Install Snowboy

   sudo apt install -y libatlas-base-dev swig python3-pyaudio sox
   sudo pip3 install voice-engine
   git clone https://github.com/Kitt-AI/snowboy.git --depth=1
   cd snowboy/swig/Python3
   make
   cd ../..
   python3 setup.py build
   sudo python3 setup.py bdist_wheel
   sudo pip3 install --no-deps dist/snowboy-*-py3-none-any.whl
   

2. Detect a keyword

   echo "
   import time
   from voice_engine.source import Source
   from voice_engine.kws import KWS
   
   
   src = Source(rate=16000, channels=1, device_name='default')
   kws = KWS(model='computer', sensitivity=0.6, verbose=True)
   src.pipeline(kws)
   
   def on_detected(keyword):
       print('\ndetected {}'.format(keyword))
   
   kws.set_callback(on_detected)
   src.pipeline_start()
   while True:
       try:
           time.sleep(1)
       except KeyboardInterrupt:
           break
   src.pipeline_stop()
   " > kws.py
   python3 kws.py