Wake Word Detection

Team members:

Yi Lin (yl182@rice.edu)

Huijia Zhu (hz73@rice.edu)

Zikuan Zhang (zz68@rice.edu)

We followed hints in Chapter 7 of the book "TinyML: Machine Learning with TensorFlow on Arduino and Ultra-Low Power Micro-Controllers" by Pete Warden and Daniel Situnayake, as well as basics in their Github page "https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/micro/examples/micro_speech#deploy-to-arduino". This is a project running on Arduino Nano 33 BLE Sense that detects key wake words in speech content. Specifically, the accuracy of detecting "yes" and "no" is improved. The functionality of detecting wake words in real-life applications is to wake up a larger program that does more important things. In this implementation, when "yes" is detected, the LED flashes green light; when "no" is detected, the LED flashes red light; when other words are detected, the LED flashes blue light. In some cases we see purple light because blue + red = purple, and that's because the time between the detected "yes" and "no" is too short. Our demo gives a short example of three colors of LED light.

The demo of R, G, B lights on LED:

A short video demo showing detection of "yes" (green), "no" (red), and other (blue)