Tags: C++Arduino Nano 33 BLE SenseQualcomm Tiny ML KitEdge ImpulseNeo Pixels

Imagine the challenges faced by those caring for loved ones with dementia. Aazami is designed to address the gap to elevate the life quality of dementia patients and their families. It records the last 10 seconds of audio, which can be replayed by saying "I forgot".

Developing Aazami involved integrating hardware and software components, including Arduino and Adafruit’s Neopixel for hardware, and Edge Impulse for Machine Learning. Honestly, we were relatively new to Machine Learning back then. We used a built-in framework rather than training the model from scratch; we initiated the process by recording various voices for voice detection. We achieved around 95% accuracy in both training and validation sets. However, due to time constraints and limited datasets, there is a possibility that the model is overfitting and biased towards a specific voiceset.

Despite its areas for improvement, Aazami won the MakeUofT Category of Most Innovative Power Efficient Hack using the Tiny ML Kit.