Verification-Based Design of a Robust EMG Wake Word

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Signal Processing

Background:

• Surface electromyography (sEMG) is used for myoelectric control and gesture recognition.
• Existing EMG gesture recognition lacks robustness for real-world single-command applications due to focus on multi-class performance.
• Current methods often use windowed classification, neglecting temporal gesture structures.

Purpose of the Study:

• To develop a robust EMG wake word system using a verification-based approach.
• To address the challenge of false activations in EMG-based gesture recognition.
• To leverage temporal structures of gestures for improved accuracy.

Main Methods:

• Proposed a verification-based approach using one-class classifiers: Support Vector Data Description, One Class-Support Vector Machine, Dynamic Time Warping (DTW), and Hidden Markov Models.
• Utilized Area Under the ROC Curve (AUC) for feature optimization.
• Evaluated performance using Equal Error Rate (EER) and AUC on a dataset of five gestures.

Main Results:

• Achieved a best Equal Error Rate (EER) of 0.04 and an Area Under the ROC Curve (AUC) of 0.98.
• The Dynamic Time Warping (DTW) scheme yielded the highest performance.
• Demonstrated superior verification performance compared to traditional window-based methods.

Conclusions:

• The proposed verification-based framework enhances the robustness of EMG wake words.
• This approach offers a promising solution for reliable EMG-based commands in interactive applications.
• The findings suggest a new direction for developing more dependable EMG interfaces.