A Wearable Silent Text Input System Using EMG and Piezoelectric Sensors

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Signal Processing

Background:

• Silent speech recognition is crucial for assistive communication and discreet input.
• Existing methods often require bulky equipment or invasive procedures.
• Wearable, non-audible systems offer a novel approach to silent text input.

Purpose of the Study:

• To develop and evaluate a wearable silent text input system using Electromyography (EMG) and piezoelectric lead zirconate titanate (PZT) sensors.
• To compare the performance of various machine learning models for classifying silent speech signals.
• To assess the system's accuracy and real-time capabilities for practical applications.

Main Methods:

• Integration of miniaturized EMG and PZT sensors into a chin-attachable wearable device.
• Acquisition of sensor data corresponding to silent articulation of English alphabet letters.
• Analysis of time and frequency domain features from sensor signals.
• Comparison of feature-based and non-feature-based machine learning models for classification.

Main Results:

• Non-feature-based machine learning models, specifically Fea-Shot Learning, demonstrated superior performance.
• The fused EMG and PZT signal approach achieved the highest accuracy (95.63%) and F1-score (95.62%).
• The system effectively captured subtle variations in muscle activity and skin vibrations associated with silent speech.

Conclusions:

• The developed wearable system provides an accurate and efficient method for silent text input.
• The combination of EMG and PZT sensors with advanced ML models shows significant potential for assistive communication.
• This technology offers a discreet and non-audible alternative for text entry in various environments.