Role of Brainwaves in Neural Speech Decoding
Neural speech decoding aims at direct decoding of speech from the brain to restore speech communication in patients with locked-in syndrome (fully paralyzed but aware). Despite the recent progress, exactly which aspects of neural activities are characterizing the decoding process is still unclear. Neural oscillations have been associated with playing a key functional role in neural information processing and thus might provide significant insight into the decoding process. Previous research has investigated a limited range of neural frequencies for decoding, usually the high-gamma oscillations (70−200Hz) in electrocorticography (ECoG) and lower-frequency waves (1−70Hz) in electroencephalography (EEG). Hence, the exact contribution of specific frequency bands is still unclear. Magnetoencephalography (MEG) is a non-invasive method for directly measuring underlying brain activity and has the temporal resolution needed to investigate the role of cortical oscillations in speech decoding, which we attempted in this study. We used three machine learning classifiers (linear discriminant analysis (LDA), support vector machine (SVM), and artificial neural network (ANN) to classify different imagined and spoken phrases for finding the role of brainwaves in speech decoding. The experimental results showed a significant contribution of low-frequency Delta oscillations (0.1−4 Hz) in decoding and the best performance was achieved when all the brainwaves were combined.