Classification of EEG Signals for Telugu Vowels

This paper explores the potential of a Brain-Computer Interface (BCI) system for recognizing Telugu vocal and sub-vocal vowels, intending to improve communication for Telugu-speaking patients with neurodegenerative disorders. We propose classifying Electroencephalogram (EEG) signals gener-ated when Telugu vowels are spoken in a vocal or subvocal manner. EEG signals collected from an 8-channel EEG device underwent preprocessing, including Butterworth bandpass filtering, to enhance quality and extract relevant features. Frequency- domain features were extracted using Power Spectral Density (PSD) analysis. For classification tasks, Random Forest and Gradient Boosting Machine algorithms were chosen due to their robustness and high accuracy in handling complex datasets. The evaluation of our classifier included performance metrics such as average testing accuracy, precision, recall, and F1 score. In three-class classification, Random Forest achieved an accuracy of 73<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> for vocal data and 80% for subvocal data. Gradient Boosting showed similar performance across both datasets. For four-class classification, Random Forest attained accuracy values of 81 % for vocal data and 71 % for subvocal data, while Gradient Boosting achieved 78% and 71 %, respectively. These results underscore the potential of machine learning in improving communication interfaces for Telugu-speaking individuals with neurodegenerative disorders.