Word Error Rate Comparison between Single and Double Radar Solutions for Silent Speech Recognition

Abstract

Silent speech recognition (SSR) is a technology that translates human speech into text without voice information. Various sensors, such as vision, electromyography, electromagnetic articulatory, and radar sensors, can be used to build an SSR system. Because the signals of radar sensors are less intuitive, radar-based SSR research is less common and remains at a basic level compared with work on other sensors. As a basic step in this research area, in this study, we attempted to determine whether single radar or double radar shows better performance for an SSR system. To this end, we estimated the word error rate (WER) of each system. The results showed that a double-radar-based SSR system produced better WER output. This means that the number of radar sensors used in SSR can potentially affect its performance. Therefore, when we create a radar-based SSR hardware platform, how many radar sensors would be ideal for its best performance must be considered.