Transformer-based neural speech decoding from surface and depth electrode signals.

This study investigates speech decoding from neural signals captured by intracranial electrodes. Most prior works can only work with electrodes on a 2D grid (i.e. Electrocorticographic (ECoG) or ECoG array) and data from a single patient. We aim to design a deep-learning model architecture that can accommodate both surface ECoG and depth (stereotactic EEG or sEEG) electrodes. The architecture should allow training on data from multiple participants with large variability in electrode placements. The model should not have subject-specific layers and the trained model should perform well on participants unseen during training.We propose a novel transformer-based model architecture named SwinTW that can work with arbitrarily positioned electrodes by leveraging their 3D locations on the cortex rather than their positions on a 2D grid. We train subject-specific models using data from a single participant and multi-subject models exploiting data from multiple participants.The subject-specific models using only low-density 8 × 8 ECoG data achieved high decoding Pearson Correlation Coefficient with ground truth spectrogram (PCC = 0.817), over= 43 participants, significantly outperforming our prior convolutional ResNet model and the 3D Swin transformer model. Incorporating additional strip, depth, and grid electrodes available in each participant (= 39) led to further improvement (PCC = 0.838). For participants with only sEEG electrodes (= 9), subject-specific models still enjoy comparable performance with an average PCC = 0.798. A single multi-subject model trained on ECoG data from 15 participants yielded comparable results (PCC = 0.837) as 15 models trained individually for these participants (PCC = 0.831). Furthermore, the multi-subject models achieved high performance on unseen participants, with an average PCC = 0.765 in leave-one-out cross-validation.The proposed SwinTW decoder enables future speech decoding approaches to utilize any electrode placement that is clinically optimal or feasible for a particular participant, including using only depth electrodes, which are more routinely implanted in chronic neurosurgical procedures. The success of the single multi-subject model when tested on participants within the training cohort demonstrates that the model architecture is capable of exploiting data from multiple participants with diverse electrode placements. The architecture's flexibility in training with both single-subject and multi-subject data, as well as grid and non-grid electrodes, ensures its broad applicability. Importantly, the generalizability of the multi-subject models in our study population suggests that a model trained using paired acoustic and neural data from multiple patients can potentially be applied to new patients with speech disability where acoustic-neural training data is not feasible.