Audio Classification on Low-Resource Microcontrollers

Recently, as the importance of using artificial intelligence in fields such as the Internet of Things (IoT) and smart factories has become prominent, research has been conducted to perform deep learning on small devices like microcontrollers. Compared to other computational devices traditionally used for deep learning, microcontrollers have very limited resources. Therefore, an optimization technique considering hardware constraints is required to perform audio classification on a microcontroller. To address this, this paper presents a method for executing a model on a microcontroller by optimizing it to fit within hardware constraints, and the process for deploying the optimized model to the device. Finally, it was demonstrated that audio classification can be successfully performed on a target microcontroller using the STM32H747I-Discovery.