The Effects of Electrode Locations on Silent Speech Recognition using High-Density sEMG

Abstract

In circumstances when silent speech is necessary, it is an attractive technology to use the surface electromyography (sEMG) signals recorded from electrodes placed on the face and neck regions for automatic speech recognition. For this technology, the electrode location is an important factor for the performance of speech recognition. However, it remains unclear how the electrode positions affect the performance, so there is no guideline for electrode placement in practical applications. Thus, the method of high-density sEMG was proposed to record sEMG signals from four electrode arrays over the facial and neck muscles. The high-density sEMG signals were utilized to analyze the effects of the electrode locations on the classification accuracies when increasing the category number of speaking tasks. The results showed that the increase in category numbers of speaking tasks would lead to the decline of classification accuracies, no matter which of the four electrode arrays were used. Meanwhile, the declining rates were the highest when using signals from the face electrodes (F-40ch), while those were the lowest when using the neck electrodes (NO-40ch and NE-40ch). There was no significant difference in the accuracies between the cases of the NO-40ch and NE-40ch electrodes. The findings indicated that the muscles on the neck region might be a more important contributor for automatic silent speech recognition. The study might provide new clues and guidelines for electrode placement when using sEMG for automatic silent speech recognition, which is important to develop a practical communication system for dysphonia.