A Super-Flexible and High-Sensitive Epidermal sEMG Electrode Patch for Silent Speech Recognition

Abstract

Surface electromyography (sEMG) signals generated by the muscles of the lower jaw contain corresponding voice information when talking. However, it's very hard to collect sEMG signals from the jaw because of not only the weakness of the signals but also the sharp skin curvature of this area. This paper reports a super-flexible and high-sensitive epidermal sEMG electrode patch to measure sEMG signals of the jaw from which the voice information can be extracted for speech recognition. This epidermal sEMG patch can be effectively attached to the jaw and the interconnection wire in the form of double wave can stretch more than 100%. From the characteristic of envelopes of the signals, it shows good consistency when the tester speak the same word and obvious difference can be detected when the tester speak different words. Four-layer wavelet decomposition is used to reduce noise in sMEG records and the eigenvectors are extracted from wavelet coefficients to characterize each trial. The experimental result shows that the average accuracy of the three pronunciations is about 71.7%. The accuracy of “hello” reaches up to 91%. In the future his epidermal sEMG electrode patch has great potential for the application of silent speech recognition.