EEG-Based Inner Speech Decoding Using Phase-Locking and Spatial Features with a Dual-Branch Deep Learning Model

Background Decoding inner speech from non-invasive electroencephalogram (EEG) signals presents a promising direction for advancing brain-computer interface technologies. However, this remains a highly challenging task because of factors such as the inherently low signal-to-noise ratio, significant inter-subject variability, and the complex, non-stationary nature of neural activity captured in EEG data. Methods This study presents a novel classification framework that combines phase-locking value and common spatial patterns for feature extraction. The dimensionality of the phase-locking features was reduced using principal component analysis. The two feature types are processed independently using recurrent Neural Networks and Deep Neural Networks and then concatenated for the final classification. The model was evaluated under both subject-dependent and subject-independent settings using a publicly available EEG dataset that includes imagined speech from five individuals in two categories word: social and numerical. Results The proposed model demonstrated strong performance on the four-class classification task in a subject-dependent (S-d) setting, with accuracies ranging from 93.75% to 100%. The average accuracy was 96.88 ± 2.46% for social words and 96.17 ± 2.43% for numerical words. Additionally, the macro F1-scores exceeded 0.94 across all cases, indicating consistent and balanced classification performance. In contrast, the subject-independent (S-Ind) evaluation yielded a considerably lower accuracy of 52.0%, underscoring the persistent challenge of cross-subject generalization in EEG-based inner speech decoding. Notably, the proposed method achieved a maximum classification accuracy of 96.88%, outperforming all previously reported approaches on the same dataset, with a performance gain of 17.59% over the best existing method. Conclusion The proposed approach shows strong performance in subject-dependent inner speech decoding and extends the evaluation to a subject-independent setting. This provides a more comprehensive and realistic assessment of model performance in EEG-based inner speech decoding. While the method shows practical potential, it also highlights the ongoing challenge of generalizing across individuals.