Study On Epidermal Electronic System For Silent Speech Recognition

Throat cancer is one of the common malignant tumors of the head and neck.About 70%of patients with throat cancer will suffer from impaired swallowing difficulty in pronunciation,or even long-term loss of sound,which greatly reduces the quality of life of patients after surgery.In recent years,studies have shown that the facial and laryngeal neck muscles correspond to different motion patterns during pronunciation.The accompanying sEMG signals contain corresponding voice information,which has excellent application prospects in speech recognition.However,most of the traditional sensors used to measure the sEMG signals are made of rigid electrodes mixed with conductive gel to combine with the skin.The device is bulky and the bonding substrate is thick.The flexible skin electronic system for silent speech recognition has become one of the effective solutions.In response to the needs of silent voice recognition and communication in patients with loss of voice,this paper has designed an epidermal sEMG electrode patch with ultra-high flexibility,which can be used to measure sEMG signals.In this paper,the epidermal sEMG electrode patch is designed with a net-like structure,and the sEMG signals are measured in the differential form.The interconnection structure is designed as the double wave form,which is protected by two layers of PI.The epidermal sEMG electrodes are fabricated by a series of MEMS processing techniques,such as photolithography patterning,wet etching,and oxygen plasma etching,which are then transferred to the polyurethane film substrate through the transfer process.In this paper,the sEMG signals are decomposed based on the db4 wavelet,and the decomposed wavelet coefficients are denoised by the threshold function.The wavelet coefficients and the AR model coefficients of the reconstructed signal are used to construct the feature vector to achieve silent speech recognition with the accuracy of 90%.Finally,the sEMG detection platform is built and the sEMG signals when the subject is silently speaking are used as the control source signal to control various human-machine interaction modes such as the entelligent car,bluetooth speech synthesis,virtual character,etc.The research contents of this article include:(1)Based on the detection principle of sEMG signals,the epidermal sEMG electrodes are designed as the differential mesh electrode,and the interconnection structure is designed as the double wave form,in which the interconnection structure is protected by two layers of PI to enhance the stretchability of the device.The epidermal sEMG electrodes are prepared through a series of MEMS processes,which are then transferred to the polyurethane film substrate through the transfer process to complete device packaging.The tensile performance testing platform is built to test the tensile properties of the epidermal sEMG electrode to verify that the designed epidermal sEMG electrodes meet the strain level of human skin.(2)Based on the principle of pronunciation,the anterior belly of the lower jaw,the buccinator of the left face and the zygomaticus of the right face are used as the measurement muscle groups in this experiment.This paper designs 5 types of action commands and 6 types of emotional commands independently as experimental paradigms to express the behavior intention and emotional demands,which is applied to carry out the silent speech sEMG experiments.Based on db4 wavelet,the signals are decomposed and denoised by the threshold function.The feature vector is constructed with the wavelet coefficients and the AR model coefficients of the reconstructed signal,and input into the LDA linear classifier for classification and recognition,the accuracy of which can reach up to to 90%.(3)The sEMG signal detection platform is built to perform hardware filtering,amplitude amplification,and analog-to-digital conversion on the initial sEMG signal.This paper creates a host computer display interface through LabVIEW program to display sEMG signal of the muscle ch annel.In addition,the sEMG signals during the silent pronunciation of the subjects are used as the control source signal to control various human-computer interaction modes to demonstrate and verify the recognition results of silent speech,such as the intelligent car,bluetooth speech synthesis,virtual characters,etc.