A novel silent speech recognition approach based on parallel inception convolutional neural network and Mel frequency...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Silent speech recognition (SSR) aims to overcome limitations of traditional automatic speech recognition (ASR) where acoustic signals are unclear or absent.
• Current SSR methods require further development for real-world applicability.
• Surface electromyography (sEMG) offers a potential alternative signal source for silent speech detection.

Purpose of the Study:

• To develop a novel and effective silent speech recognition framework utilizing sEMG signals.
• To introduce and implement a new deep learning architecture, the Parallel Inception Convolutional Neural Network (PICNN), for sEMG-based SSR.
• To create a comprehensive dataset for silent speech recognition, focusing on daily life assistance demands.

Main Methods:

• A novel deep learning architecture, Parallel Inception Convolutional Neural Network (PICNN), was designed, processing six channels of sEMG data simultaneously.
• Mel Frequency Spectral Coefficients (MFSCs) were utilized for the first time to extract speech-related features from sEMG signals.
• A 100-class dataset was generated, encompassing demands relevant to elderly and disabled individuals.

Main Results:

• The proposed sEMG-based SSR system achieved a peak recognition accuracy of 90.76% across 28 subjects.
• The PICNN architecture outperformed existing state-of-the-art machine learning and deep learning algorithms in silent speech recognition.
• Subject-based transfer learning demonstrated effectiveness in enhancing cross-subject recognition capabilities.

Conclusions:

• The developed sEMG-based silent speech recognition system demonstrates high accuracy and robust performance.
• The novel PICNN architecture and MFSC feature extraction offer a significant advancement in the field of silent speech recognition.
• This technology holds considerable potential for practical applications, particularly in assistive technologies for individuals with communication impairments.