Embedded machine learning: Without Edge impulse

Abstract

This paper presents an implementation of embedded machine learning using TensorFlow Lite Micro on two microcontroller boards, namely, KL25Z and Nana BLE. The implementation aims to recognize a wake word using a convolutional neural network (CNN). The paper details the design and implementation of the CNN architecture and the training and deployment process using TensorFlow Lite Micro. The results of the implementation show that the training of the CNN was successful and achieved high accuracy in wake word detection. However, deploying the trained model on the microcontroller boards was unsuccessful.