ML Doggie Door

Machine Learning Bark and Keyword Recognition
A homemade doggie door with electric lock
Project Overview
This project was focused on TinyML, an emerging field focusing on utilizing machine learning in microcontrollers. I gathered, and prepared the data to train a voice recognition TensorFlow Lite model to recognize and differentiate between the sound of a bark, activation word, silence, and random noises that could be found around a home. The goal was to allow home owners to let their pets out without leaving an entry point open into their home.

As you can see in the Confusion Matrix below, the next iteration can be improved with better data and training settings.
There are two options for use for this project. In the first, upon recognition of a bark, the system notifies the dog owner that the dog is barking in the vicinity of the door. The owner can then say "go ahead" and the door will unlock, allowing the dog out. Alternatively, the owner can train their dog to bark at the door and have it unlock automatically. Not all dogs are trained equally or equally trainable.

Future iterations could also be trained for any other pet that makes noise. The addition of two simple components, like a time of flight sensor on either side of the door, could also be added to track the direction of movement. This will let you know whether your pet is in or out of the house, a common question for pet owners who let their pets roam freely.
Role
Solo Project
Dec 2021