EEG-Based Envisioned Speech Recognition System Using Spectral Graph Wavelet Transform

This paper presents a new framework for the recognition of envisioned speech from electroencephalogram (EEG) signal. The proposed framework consists of preprocessing for dividing the signals into blocks, spectral graph wavelet transform (SGWT), feature smoothing using the moving average filter, and classifiers (extra trees (ET), random forest (RF), and k-nearest neighbors (KNN)) for identifying three tasks such as digits, characters, and objects. The performance of the proposed method is evaluated on a publicly available database which consists of three classification tasks namely, digits, characters, and objects tasks. The SGWT-based method with ET classifier had the highest accuracies of 92.4%, 92.7%, and 92.3% for digits, character, and objects tasks, respectively that outperforms other SGWT-based methods using the RF and KNN. Evaluation results show that the class-wise accuracies are better than the state-of-the-art methodologies. The proposed EEG-based framework for recognition of envisioned speech can enable seamless brain-computer interfaces (BCIs) for communication of people having speech impairments and can control the devices using envisioned speech in human-machine interaction (HCI) applications.