Exploring TinyML Frameworks for Small-Footprint Keyword Spotting: A Concise Overview

Keyword spotting with a Small Footprint (SF-KWS) has gained popularity in today's landscape of smart voice-activated devices, smartphones, and IoT applications. This surge is attributed to the advancements in Deep Learning, enabling the identification of predefined words or keywords from a continuous stream of words. To implement the SF-KWS model on edge devices with low power and limited memory in real-world scenarios, a efficient TinyML framework is essential. In this study, we explore six distinct categories of techniques namely, Network Architecture, Learning Techniques, Model Compression, Attention Awareness Architecture, Feature Optimization, and Neural Network Search, which are suitable for developing an SF-KWS system. We experimented with various popular architectures using available TinyML frameworks throughout our investigation. Our findings reveal that the Tensorflow Lite framework can compress a model by approximately 69%, maintaining acceptable accuracy. Additionally, though the Edge Impulse framework reduces memory footprint significantly, are approximately 13.3% less accurate than MicroNet-S.