Enhancing Word Recognition in Imagined Speech Using Non-Invasive EEG for Improved BCI Applications

This study investigates the recognition and classification of words in imagined speech using non-invasive EEG signals. A method combining a generator, discriminator, and classifier to reconstruct and classify Mel-spectrograms is proposed, while domain adaptation techniques and optimization models are employed to enhance the accuracy and reliability of decoding. The results demonstrate the feasibility of this non-invasive method, offering significant implications for developing practical Brain-computer Interface (BCI) communication tools, particularly for individuals with communication impairments, and expanding the applicability of BCI across diverse fields.