Abstract :

Translating Human Brain Activity Into Coherent Text Using Electroencephalography (EEG) Is A Transformative Frontier In Brain-Computer Interfaces (BCI), Offering A Communication Lifeline For Individuals With Speech And Motor Impairments. While Recent Advancements, Such As The Thought2Text Framework, Have Utilized Multi-stage Training And Large Language Models (LLMs) To Bridge This Gap, They Often Introduce High Computational Overhead And Complex Multimodal Dependencies. This Project Presents An Optimized, End-toend Transformer-based Approach That Streamlines The Decoding Of Continuous EEG Signals Into Natural Language. Our Methodology Extends The Current State-of-the-art By Replacing Complex Multi-stage Alignment With A Direct Temporal-Spatial Mapping Technique, Reshaping Raw EEG Vectors Into A Structured $105 Imes 6$ Sequence To Better Leverage Self-attention Mechanisms. We Implement A Custom Transformer Encoder Architecture Equipped With Learnable Positional Embeddings And A Specialized Linear Word-projection Head. To Enhance Translation Fluency And Mitigate The "noise" Inherent In Non-stationary Brain Signals, We Adopt A Frequency-constrained Vocabulary Strategy Targeting The Top 1,000 Most Common Tokens. Experimental Results Demonstrate That Our Model Achieves A Validation Accuracy Of ~23.95% On Complex Sentence Reconstruction, Successfully Capturing Key Entities And Semantic Structures From Raw EEG Data. By Achieving Competitive Performance Without The Need For Secondary Visual Stimuli Or Massive LLM Backbones, This Approach Offers A More Computationally Efficient And Portable Solution For Real-time "thought-to-text" Translation, Addressing The Scalability Challenges Highlighted In Contemporary BCI Literature. Keywords: EEG-to-Text, Brain-Computer Interface (BCI), Transformer Encoder, Signal Decoding, Deep Learning, Natural Language Processing.

Published:

31-8-2026

Issue:

Vol. 26 No. 8 (2026)

Page Nos:

1127 - 1132

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite