EEG-based classification of bilingual unspoken speech using ANN

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Interpreting unspoken speech via electroencephalography (EEG) offers therapeutic potential for speech disorders and locked-in syndrome.
• Existing brain-computer interface (BCI) methods using single-language EEG classifiers achieve limited accuracy (40-60%), hindering practical application.
• Multilingual societies necessitate BCI systems capable of handling multiple languages for broader utility.

Purpose of the Study:

• To develop a novel bilingual approach for interpreting unspoken speech using EEG.
• To improve the accuracy and practical applicability of EEG-based BCI systems.
• To investigate the efficacy of combining Hindi and English speech data for enhanced classification.

Main Methods:

• Collected EEG data from 5 bilingual subjects responding to 'Yes'/'No' questions in Hindi and English.
• Utilized electroencephalography (EEG) sensors in brain regions critical for language processing and decision-making.
• Applied Principal Component Analysis (PCA) for dimensionality reduction, followed by Support Vector Machine (SVM), Random Forest (RF), AdaBoost (AB), and Artificial Neural Networks (ANN) for classification.

Main Results:

• Artificial Neural Networks (ANN) achieved the highest accuracies: 85.20% for decision classification and 92.18% for language classification.
• The combined bilingual approach yielded an overall speech classification accuracy of 75.38%.
• The study demonstrated significant improvements over previous single-language EEG-based speech interpretation methods.

Conclusions:

• Bilingual EEG-based speech interpretation significantly enhances classification accuracy compared to single-language models.
• The developed approach holds promise for advancing brain-computer interfaces (BCI) for individuals with communication impairments.
• Future research should explore larger datasets and diverse linguistic combinations to further refine BCI technology.