Source-linked AI summary
Soft Active Electromyography Interface for Machine Learning-Enabled Silent Speech Recognition
Yuta Kurotaki, Shusuke Yamakoshi, Reitaro Yoshida, Yutaka Isoda, Tamami Takano, Yuji Isano, Yusuke Miyake, Kentaro Kuribayashi, Hiroki Ota
TL;DR
The study develops and fabricates a soft EMG interface with fingertip electrodes and liquid-metal wiring, then applies recognized words to real-time drone control. Electrode–skin impedance was characterized through eight independent measurements.
Problem
The study addresses the need to acquire and apply signals from a soft wearable EMG interface for silent-speech control.
Method
A soft EMG device was fabricated by integrating fingertip electrodes, liquid-metal wiring, an amplification circuit, and elastomer encapsulation, then connected to a drone-control program.
Results
The dry fingertip electrode’s impedance spectrum was measured over 10 Hz to 10 kHz using n = 8 independent forearm-skin measurements.
Takeaways & Limitations
Recognized words were transmitted via HTTP and used to control drone movements within a room in real time.
Takeaways & Limitations
The participant evidence was collected under an ethics-approved protocol involving informed-consenting participants.
Abstract
from arXiv · showhide
Silent speech recognition (SSR) provides an alternative communication pathway in the absence of audible speech. However, conventional approaches are limited by the need for constant facial attachment, privacy concerns, and unstable signal acquisition. Here, we propose a soft, active electromyography (EMG) interface that enables word-level SSR using machine learning. Worn on the hand, the device uses a fingertip electrode that can be positioned near the lips to acquire EMG signals only when needed. The interface integrates liquid metal (LM) interconnects, transparent flexible printed circuit (FPC) electrodes, and elastomer encapsulation to ensure high mechanical stability during finger motion. A deep neural network trained on these stable signals achieved a mean accuracy of 97.2 $\pm$ 1.3% across three subjects in classifying a 30-word vocabulary, demonstrating robust linguistic discrimination. Furthermore, real-time drone control validates the practicality of this approach in noisy and privacy-sensitive environments where conventional voice recognition fails. This study highlights the potential of soft, wearable EMG systems as secure and intuitive human-machine interfaces.
Preparation and wiring of LM paste
LM paste was prepared by dispersing nickel powder in Galinstan, followed by sonication and overnight air exposure to promote oxidation. An alternative oxidized paste was produced by stirring Galinstan under ambient conditions.
- Preparation: Nickel powder (3–7 µm) was dispersed in Galinstan at mass ratios of 2% and 5%.The nickel powder was sourced from Alfa Aesar Co., and Galinstan from Maruya.com.
- Preparation: 6 kJ of sonication energy was applied using a 30% duty ratio to the nickel–Galinstan mixture.Sonication used an ultrasonic probe (SFX 550, BRANSON).
- Oxidation: Overnight air exposure promoted oxidation of the sonicated liquid-metal paste.This produced oxidized LM paste after the mixture was exposed to air overnight.
- Oxidation: 750 rpm stirring for 60 min under ambient conditions induced oxidation in Galinstan.An Azone stirrer was used to prepare this alternative oxidized LM paste.
Fabrication of silent speech device
The device was fabricated by molding and curing Ecoflex elastomer, attaching the EMG electronics and fingertip electrodes, and forming LM wiring through a laser-cut polyimide stencil. A Python-based control program connected the device to a Tello drone via Wi‑Fi for silent-speech command execution.
- Device fabrication: Ecoflex 00-20 silicone rubber was cast in a 3D-printed mold and oven-cured at a 1:1 A:B weight ratio.The elastomer formed the device body.
- Device fabrication: The EMG amplification circuit board and fingertip electrodes were fixed to the cured elastomer using Sil-Poxy adhesive.Both components were attached using the same adhesive method.
- Device fabrication: A laser-cut polyimide film served as a stencil for applying liquid-metal paste to form wiring connections.The LM paste was applied through the mask, which was removed afterward.
- Drone operation application: A Python control program transmitted recognized silent-speech words via HTTP to a Tello drone connected over Wi‑Fi.The recognized words were also displayed in a web browser while the drone executed movements indoors.
Ethical approval and participant consent
Participants provided informed consent after receiving a full explanation of the study’s purpose and procedures, and the protocol received ethics approval from Yokohama National University’s engineering science committee.
- All participants provided informed consent after receiving a full explanation of the study’s purpose and procedures.
- The research protocol was approved by the Ethics Committee of the Yokohama National University Graduate School of Engineering Science (No. 2020-16; approved February 12, 2021).
Code availability
The study’s underlying code and training/validation datasets are planned for release in a publicly accessible repository, with the persistent URL to be provided upon publication.
- Code availability: The underlying code and training/validation datasets are planned for release in a publicly accessible repository.The repository will have a persistent URL, and the link will be provided at publication.