Brain-Computer Interface System Based on Common Spatial Patterns for Inner Speech Recognition from Electroencephalography Signal by Using a Convolutional Neural Networks

Inner speech recognition is a modern advancement in Brain computer interfaces (BCI) that facilitates a communication between the computer and the brain in a direct way. It is particularly beneficial for individuals who face communication issues due to speech disability. In this research, an end-to-end brain-computer interface system was developed that uses Electroencephalography (EEG) signals to recognize inner speech straight from the brain. The study utilized an open-access dataset presented on four Spanish words from ten participants. The proposed system involved preprocessing to clean and enhance the signal. Common Spatial Pattern (CSP) was utilized to extract features to improve the discrimination between the selected four words. A Convolutional Neural Networks (CNNs) deep learning model was used to enhance the inner speech recognition performance from spatial features. The result shows that the model can decode the inner speech with an average accuracy of 81.40% for the un-seeing dataset and 89.80% for the entire dataset, indicating that the proposed method outperforms previous works that used the same dataset.