Unraveling the Secrets of Inner Speech from Electroencephalograph with Spatial Features and Convolutional Neural Networks

カテゴリ: 研究会(論文単位)

論文No: CMN24022

グループ名: 【C】電子・情報・システム部門 通信研究会

発行日: 2024/03/25

タイトル(英語): Unraveling the Secrets of Inner Speech from Electroencephalograph with Spatial Features and Convolutional Neural Networks

著者名: Abdalla Hussna Elnoor ( University Putra Malaysia, Selangor, 43400, Malaysia \ Sudan International University, Khartoum, Sudan ), AL-HADDAD SYED ABDUL RAHMAN( University Putra Malaysia, Selangor, 43400, Malaysia. ),Basri Hamidon (University Putra Malaysi

著者名(英語): Hussna Elnoor Abdalla( University Putra Malaysia, Selangor, 43400, Malaysia \ Sudan International University, Khartoum, Sudan ),SYED ABDUL RAHMAN AL-HADDAD( University Putra Malaysia, Selangor, 43400, Malaysia. ),Hamidon Basri (University Putra Malaysi

キーワード: Brain Computer Interface|Inner Speech |Electroencephalogram |Common Spatial Pattern|Convolution Neural Networks |Deep Learning |Brain Computer Interface|Inner Speech |Electroencephalogram |Common Spatial Pattern|Convolution Neural Networks |Deep Learning

要約(日本語): Brain-computer interfaces (BCI) allow for direct brain-to-computer communication. The brain produces speech using oral articulators, which then convert it to sound. BCI is the best option to address communication problems with individuals who are disabled and unable to talk because they have lost their oral articulators. This work used non-invasive neural signal Electroencephalography which has well investigated the decoding of inner speech from the brain presented in four Spanish words collected from ten participants. Following preprocessing, the spatial features were retrieved using a Common Spatial Pattern (CSP) to improve the spatial resolution from the EEG signals and then fed into a one-dimensional convolutional neural network-based deep learning model. The results showed the model's ability to decode the inner speech with an average accuracy of 83.50% for the un-seeing dataset and 91.50% for the entire dataset.

要約(英語): Brain-computer interfaces (BCI) allow for direct brain-to-computer communication. The brain produces speech using oral articulators, which then convert it to sound. BCI is the best option to address communication problems with individuals who are disabled and unable to talk because they have lost their oral articulators. This work used non-invasive neural signal Electroencephalography which has well investigated the decoding of inner speech from the brain presented in four Spanish words collected from ten participants. Following preprocessing, the spatial features were retrieved using a Common Spatial Pattern (CSP) to improve the spatial resolution from the EEG signals and then fed into a one-dimensional convolutional neural network-based deep learning model. The results showed the model's ability to decode the inner speech with an average accuracy of 83.50% for the un-seeing dataset and 91.50% for the entire dataset.

本誌: 2024年3月28日-2024年3月29日通信研究会

本誌掲載ページ: 31-36 p

原稿種別: 英語

PDFファイルサイズ: 1,216 Kバイト