Evaluation of Voice Recognition Platforms and Methods for Edge AI Devices

AI-driven voice recognition technologies have an important application area for remote control of devices ranging from smart home appliances, automotive systems to consumer electronics and industrial automation. Thus, many devices are being designed to have AI based voice processing capabilities. However, AI models need to be implemented in edge devices for some applications that have no access for cloud computing. So, it is very critical to optimize the resource usage as well as improving the model due to hardware limitations. On this issue, this paper presents implementation and comparative analysis of AI based voice recognition methods for edge devices via different MCU platforms that we named as Platform 1, Platform 2, and Platform 3. In this context, phoneme-based and word-based voice recognition models with related platforms were evaluated to observe recognition performance and CPU/memory usage. According to the results, Platform 1 with phoneme-based model showed best accuracy performance while Platform 3 with phoneme-based model provided the lowest memory usage in our experimental scenarios.