Electromyogram‐Based Lip‐Reading via Unobtrusive Dry Electrodes and Machine Learning Methods

Abstract

The Stony Brook University provided guidelines for study procedures. Informed consent was obtained from all participants. After data acquisition, the data were segmented into every single trial, and features were calculated. LDA/SVM training programs were applied to get the classification results. Figures 5 and 6 and Figures S14-S21, Supporting Information summarize the recognition results of all four subjects. The results of 11 words from different viseme groups using the LDA and the SVM were 94.8% ± 0.035 and 90.5% ± 0.061, respectively. The results for nine pairs of similar words were 86.8% ± 0.053 and 82.6% ± 0.056 when using the LDA and the SVM, respectively. The recognition accuracy was represented as mean ± standard derivation. Matlab was used as the software for the statistical analysis.

Decoding silent speech commands from articulatory movements through soft magnetic skin and machine learning

“…This speech recognition accuracy is comparable to the previous system based on 8-channel EMG sensors placed around lips and on the neck and tested using the same list of words. 5 This demonstrates that the selected sensing area (the ramus area near the ear) contains rich information of speech articulation, comparable to the commonly used EMG methods. In despite of similar accuracies, only a single magnetic skin is needed in this work and the sensor placement location is much more socially acceptable.…”

“…In addition to phonemes, a challenging list of words containing words from eleven visemes groups, 5 as discussed in Section 2.3, was also selected to test the developed silent speech interface. Each word was repeated fifty times.…”

“…On the other hand, environmental factors such as noisy surroundings (e.g., acoustically harsh workplaces, crowded gatherings, or background noise from televisions), situations requiring quiet or privacy (e.g., hospitals, public areas, or private communications), and environments lacking an acoustic medium (e.g., underwater or in the space) often impose limitations on voice-based speech communication. 5,6 In this study, we introduce a ground-breaking concept in the field of silent speech interfaces. Central to our concept is the utilization of a single soft magnetic skin discreetly positioned in the ramus-temporal junction area, which enables socially acceptable silent speech recognition through precise decoding of articulatory movements.…”

“…35 Though effective, it is an invasive method and requires wiring to connect EPG electrodes in the mouth to an external circuit. EMG-based systems 5,[36][37][38][39][40][41] are very promising, and researchers have developed soft conformal dry EMG electrodes 42,43 to improve their signal quality and long-term wearability. However, EMG-based interfaces often require multiple electrodes placed on the face or the neck, increasing system complexity and reducing user acceptance.…”

Bio-Inspired Untethered Robot-Sensor Platform for Minimally Invasive Biomedical Sensing

“…Close monitoring of health parameters facilitates early diagnosis and provides effective therapeutic guidance. − The conventional catheter or probe-based monitoring is uncomfortable and inconvenient, making it infeasible for frequent monitoring. , When used for continuous monitoring, the procedures are not risk-free and can result in inaccurate readings if patients’ daily activities are affected by the presence of the invasive probe . Recently developed implantable sensors, used as an alternative, often require invasive procedures to place the sensors in desired locations. ,− Therefore, a minimally invasive method of biomedical monitoring is in urgent demand.…”