Wake World Detection - TinyML

Introduction and Background

We use some Intelligent voice assistant in our life like Siri Alexa. So this project is to do the wake word detection, we use the model which was trained on a dataset called the Speech Commands dataset. This con‐ sits of 65, 000 one-second-long utterances of 30 short words, crowdsourced online. I choose word "backward" and "forward" to detect this project.

Steps

First, I need to train data get the result below:

Second, we need to deploy to microcontroller. Convert the TensorFlow Lite model into a C source file that can be loaded by TensorFlow Lite for Microcontrollers.

Third, modify the code a little bit

micro_features_micro_model_settings.cpp

arduino_command_responder.cpp

Result