A Variable-speed Silent Speech Recognition Method based on Surface Electromyography Signal

In the practical implementation of silent speech recognition based on surface electromyography (sEMG) signal, the change in the subjects' speech speed will affect the recognition performance. To mitigate this effect, a silent speech recognition method based on dynamic time warping (DTW) algorithm is proposed in this paper. Specifically, the facial sEMG signals of 22 Chinese words are collected, the energy threshold method is used to detect the active segment, the root mean square feature is extracted, and finally the DTW algorithm is used for recognition. The experimental results show that the method is robust to silent speech recognition tasks with different speech speeds. The average accuracy of using the DTW algorithm to classify uniform-speed words is 94.55%, and that of variable-speed words is 71.82%. In addition, the proposed method is suitable for few samples learning, which means it can quickly adapt to new tasks and individuals. These findings provide a novel way for the practical application of sEMG based silent speech recognition.