A Hybrid EEG-Based Approach for Classifying Imagined Speech Using Reactive Auditory Signals

カテゴリ：部門大会

論文No：TC2-1-1

グループ名：【C】2025年電気学会電子・情報・システム部門大会

発行日：2025/8/20

タイトル(英語)：A Hybrid EEG-Based Approach for Classifying Imagined Speech Using Reactive Auditory Signals

著者名：Zhang Zhuohao（東京科学大学）,Pangpong Phurin（東京科学大学）,Connelly Akima（東京科学大学）,Yagi Tohru（東京科学大学）

著者名(英語)： Zhuohao Zhang (Institute of Science Tokyo),Phurin Pangpong (Institute of Science Tokyo),Akima Connelly (Institute of Science Tokyo),Tohru Yagi (Institute of Science Tokyo)

キーワード：Active BCI,sound imagery,inner speech,electroencephalography (EEG),Event-related spectral perturbation (ERSP)locked-in syndrome (LIS)

要約(日本語)：Active BCIs allow system control via spontaneous thought but pose challenges in signal processing. Since spontaneous signals are weak, training models involve great user effort in repetitive imagery tasks, causing fatigue and reducing user engagement. These challenges hinder the widespread use of BCIs. To address those issues, we propose using listening for model training. We examine whether listening captures key features for imagery decoding by comparing brain activity during auditory perception and imagery. Results show similar neural patterns between two tasks, and models trained on listening data achieve comparable classification accuracy on imagery tasks, suggesting listening is an effective pre-training strategy for active BCIs, reducing user fatigue and improving system usability.

本誌掲載ページ：27-33p

原稿種別：英語

PDFファイルサイズ：1,188Kバイト