An efficient deep learning approach for automatic speech recognition using EEG signals.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) signals offer potential for human-machine interaction but face challenges in speech recognition due to signal noise and complexity.
• Accurate speaker identification from EEG is crucial for advancing brain-computer interfaces (BCIs) and assistive technologies.

Purpose of the Study:

• To develop an Efficient Deep Learning Approach (EDLA) for robust speaker identification using EEG signals.
• To integrate the Gannet Optimization Algorithm (GOA) with an Elman Recurrent Neural Network (ERNN) for improved EEG-based speaker recognition.

Main Methods:

• EEG data preprocessing using a Savitzky-Golay filter.
• Recursive feature elimination for optimal feature selection.
• Implementation of the Gannet Optimization Algorithm (GOA) and Elman Recurrent Neural Network (ERNN) for speaker identification.

Main Results:

• The proposed EDLA achieved a speaker identification accuracy of 95.2% on the Kara One dataset.
• EDLA demonstrated superior performance compared to existing baseline methods.
• The framework effectively addresses noise and complexity inherent in EEG signals for speech recognition.

Conclusions:

• The EDLA framework represents a significant advancement in EEG-based speaker identification.
• This approach holds promise for enhancing BCIs and developing assistive technologies for individuals with speech impairments.
• The integration of GOA and ERNN offers a powerful solution for complex neural signal processing.