BINGO’s paper was accepted at EUSIPCO, the premier European conference on signal processing where a novel deep learning architecture tailored to inner speech decoding from EEG signals was presented. Recent attempts to decipher imagined speech from EEG signals deploy Convolutional Neural Network (CNN) architectures such as shallow Conv Net, deep Conv Net and EEGNet while others use Cross-Covariance (CCV) matrices and Riemannian Geometry.
By combining the best of the two worlds, the proposed architecture combines EEGNet with CCV matrices, extracting discriminative features from the latter with the use of bilinear transformations as proposed in the SPDNet architecture. In essence, our approach calculates CCV matrices that can capture the brain connectivity structure that underpins the imagined speech paradigm at various frequency bands. The motivation for employing this particular architecture for the task at hand is related to the “dual stream model” according to which several brain regions are involved and interconnected during speech formulation and understanding. Our method is validated on two publicly available datasets and exhibits on par with State-of-the-Art performance, while substantially surpassing EEGNet performance on both datasets.
| More information: Rousis, G., Kalaganis, F. P., Nikolopoulos, S., Kompatsiaris, I., & Petrantonakis, P. C. Combining EEGNet with SPDNet towards an end-to-end architecture for imagined speech decoding. ISBN: 978-9-4645-9361-7 |
| Venue information: EUSIPCO 2024 |
| Article Source: EURASIP |