Decoding Imagined Speech From EEG Using Transfer Learning

We present a transfer learning-based approach for decoding imagined speech from electroencephalogram (EEG). Features are extracted simultaneously multiple EEG channels, rather than separately individual channels. This helps in capturing the interrelationships between cortical regions. To alleviate problem of lack enough data training deep networks, sliding window-based augmentation is performed. Mean phase coherence and magnitude-squared coherence, two popular measures used connectivity analysis, as features. These features compactly arranged, exploiting their symmetry, to obtain three dimensional “image-like” representation. The dimensions this matrix correspond alpha, beta gamma frequency bands. A network with ResNet50 base model classifying prompts. proposed method tested on publicly available ASU dataset EEG, comprising four different types accuracy prompt varies minimum 79.7% vowels maximum 95.5% short-long words across various subjects. accuracies obtained better state-of-the-art methods, technique good prompts complexities.