STMicroelectronics and AWS (Amazon Web Services) have developed the AWS STM32 ML at the Edge Accelerator. The application demo leverages the B-U585I-IOT02A Discovery Kit, the STM32Cube.AI developer cloud, and AWS solutions to run an audio classification model on an STM32U5 microcontroller. The application demo shows how ST technologies such as Model Zoo and Board Farm can extend AI at the edge.
We start with YAMNet-256, an audio event detection model from the ST Model Zoo. This model uses the B-U585-IOT2A discovery kit and X-CUBE-AWS, an extension pack that integrates FreeRTOS and AWS IoT Core for seamless cloud connectivity. The architecture supports the entire MLOps process, with the machine learning stack handling data processing, model training, and evaluation. The IoT stack handles automatic flashing of devices using over-the-air (OTA) updates, ensuring that all devices are running the latest and secure firmware.
The Pipeline stack orchestrates a CI/CD (Continuous Integration/Continuous Delivery) workflow to ensure work is always updated and code is optimized, allowing developers to automate the deployment of ML and IoT stacks to address the entire development lifecycle.
To monitor the relevant devices and visualize the data, we used Amazon Grafana to create a dynamic and interactive dashboard for real-time monitoring and analytics. The demo features a Yamnet audio classification model optimized for an STM32 MCU running an audio event detector. It can distinguish between a variety of noises, from dogs barking to birds chirping to people coughing and sneezing, and other sounds.
