Left/right contributions of articulatory muscles in speech recognition using high-density surface electromyography

Speaking process is one of the most important activities for human communication and it is controlled by the coordination of muscles on the face and neck. Surface electromyography (sEMG), an effective tool for detecting the electrophysiological features of muscles, was used frequently for speech recognition. Generally, the electrode position is one of the main factors impacting the classification accuracies of speech recognition. However, current sEMG-based speech recognition approaches usually depend on the electrodes chosen by experience, and whether electrodes of both left and right side should be chosen remains unclear. In this study, high-density (HD) sEMG signals were symmetrically recorded from 120 electrodes placed on both sides of the facial and neck muscles across eight normal-speaking subjects, when they were speaking five Chinese and English words, respectively. The support vector machine (SVM) classifier were applied on four time-domain features extracted from the HD sEMG signals for speech classification. The results showed that classification accuracies in speech recognition could be larger than 85% in average, and using information from the left side of the neck muscles had no significant difference from using the right side. Meanwhile, the individual difference played an import role in the left/right contributions of the facial muscles. The pilot study suggests that the left/right contributions of the articulatory muscles should be different between the neck and face, which might provide useful clue to reduce the electrode number and to select the best location of channels for speech recognition.