I am from a family of four and in families you have to make sacrifices, but long showers seem non negotiable for EVERYONE. To stop this chaos before school/work you can of course use an alarm or a timer but that makes it so easy to cheat. This is something along the lines of a modified shower alarm made with machine learning. As Wio Terminal has MIC, Buzzer and Display, it was the perfect tool for this project. While I tested out both the loudness sensor and the MIC, I decided to use the MIC as it has a higher rate of accuracy.
Everything in machine learning starts with data, and the very first step of building a machine learning model is determining what data you actually need and collecting it. For this model we know we want to obtain raw audio data of typical bathroom activities (brushing teeth, sink on/off, walking around). This took me long time as I needed o gather a variety of data and sounds for this to work. Then I gathered as much data as I could of the actual shower.
Once the data was collected, I used edgeimpulse to train it so the machine would be able to differentiate between the shower and other noise. The system will check the shower and once it find out that the shower is running, the clock time starts after 8 min, the buzzer will trigger.
I could get only 83.3% accuracy so in the future I will see if I need to get additional data or change any parameter to increase it. I would also like to extend this basic idea to other appliances such as the AC, the washing machine and more!
Comments