A silent speech recognition method based on facial and neck surface electromyography

Abstract

The invention discloses a silent speech recognition method based on surface electromyography of the face and neck, which performs data preprocessing and feature extraction on surface electromyography signals collected by high-density electrode arrays and discrete electrodes to obtain high-density sEMG image sets and channel sparseness sEMG image set to construct the source domain database and the target domain database; then use the source domain database to train the word classification deep neural network and use transfer learning in the target domain database to complete the calibration of the network; if the test user is silent under the discrete electrode input Expressing words, the calibrated network performs word classification and unvoiced speech recognition. The invention takes into account the high-density electrode array's ability to capture rich muscle activation pattern information and the lightness and wearability of discrete electrodes, and has certain robustness to slight electrode offset and cross-user conditions, and improves the performance of discrete electrode input. The performance of silent speech recognition provides a new idea for the silent speech recognition method.

Description

technical field [0001] The invention belongs to the fields of biological signal processing, machine learning and intelligent control, and specifically relates to a silent voice recognition method based on facial and neck surface electromyography. Background technique [0002] Voice interaction is one of the most natural and direct ways for people to interact, because the voice signal contains information such as the emotion and intention that the speaker wants to express. Automatic speech recognition (automatic speech recognition, ASR) refers to the computer to analyze and understand the collected speech signal, and convert it into text or other forms of information. ASR plays a vital role in human-computer interaction, but it also has limitations in special scenarios, such as high-noise backgrounds, people with vocal impairments, and private input environments. Therefore, how to overcome these difficulties in practical applications has always been a hot topic in the resear...

AI Technical Summary

Problems solved by technology

Transfer learning methods based on deep neural networks can adapt EMG pattern classifiers to current electrode positions or new user domains, but also complicate the use of EMG control and increase the training burden for users

Moreover, most of these studies focus on different tasks under the same measurement electrode conditions, and the migration methods across electrode domains are hardly considered and discussed.

Images