Parallel-Inception CNN Approach for Facial sEMG based Silent Speech Recognition

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Elderly individuals often face communication challenges.
• Existing human-machine interaction methods may not be suitable for all users.
• Silent speech recognition offers a potential solution for enhanced communication.

Purpose of the Study:

• To develop a novel human-machine interaction platform for the elderly.
• To create a facial electromyography-based silent speech recognition system.
• To evaluate the performance of a new deep learning architecture for this application.

Main Methods:

• A Parallel-Inception Convolutional Neural Network (PICNN) deep learning architecture was proposed.
• Log Mel frequency spectral coefficients (MFSC) were used for feature extraction.
• A 100-class dataset of daily life demands was created for training and testing.

Main Results:

• The proposed PICNN framework achieved a highest recognition accuracy of 88.44%.
• This performance exceeded state-of-the-art algorithms like CNN, VGGNet, and Inception CNN.
• The system demonstrated significant improvements over existing methods.

Conclusions:

• The developed silent speech recognition system shows promise for reliable communication.
• This technology can expand the application scope of speech recognition.
• The system offers a viable communication channel for the elderly and those in need.