Comparison of Contributions between Facial and Neck Muscles for Speech Recognition Using High-Density surface Electromyography

Speaking is an important way for human beings to communicate each other. Generally, voice signals are used for speech recognition, which is easily affected by environmental interference. Additionally, surface electromyography (sEMG) of articulatory muscles has been proposed in previous studies to enable speech recognition, which is insensitive to noisy environments. However, it remains unclear what are the contributions of facial and neck muscles for speech recognition, which would be vital for selecting locations of sEMG recording electrodes. In this study, the high-density (HD) sEMG technique was proposed to explore the major articulatory muscles contributed to speaking. The HD sEMG signals were acquired from four subjects by surface electrodes over the face and neck during speaking five Chinese daily phrases, from which four features (mean absolute value, waveform length, number of zero crossing, and slope sign change) were extracted. Then a linear-discriminant-analysis classifier was built by the sEMG features for speech recognition. The primary results showed that the sEMG and RMS waveforms illustrated obvious difference when speaking different Chinese phrases. And the classification accuracy using signals from the neck was higher than that from the facial muscles, whereas the accuracy was increased by using the whole facial and neck muscles. Our pilot results revealed that the facial and neck muscles were both contributed to the speech recognition while the neck muscles were more crucial than the facial muscles during speaking. This pilot study may suggest that the HD sEMG might pave a way to find the major muscles of speech recognition.