Design of Custom Keyword Recognition using Edge Impulse on Arduino Nano 33 BLE Sense

Generally neural networks will get trained and executed using large processing units (TPU's -Tensor Processing units) and requires large amount of computational power. Many hardware devices connected to cloud server to get required hardware and computational power for the purpose of execution of neural networks. The intent of this project is to build a neural network which is trained using real time custom keyword dataset and executed in a microcontroller (Arduino Nano 33 BLE sense) which consists of one megabyte of flash memory and 256 Kilobytes of RAM and to control the built-in led present on microcontroller with our custom keyword. The device works with TensorFlow module for recognizing human voice based on classification using Deep Learning. The objective of custom keyword voice recognition is to control the electronic devices with our custom keyword.