This study explores the neural mechanisms underlying imagined speech and its potential for brain-computer interface (BCI) applications, particularly for individuals with severe communication impairments. Using EEG and a novel deep learning model, we aimed to decode linguistic components from imagined speech, focusing on vowels and semantic categories. Our model, which integrates convolutional and recurrent neural network architectures, was validated through the classification of imagined speech, overt speech, and silence, achieving a significant accuracy of 81.76%. High accuracy was also obtained in vowel classification during imagined speech (92.06%) and in semantic categorization (84.88%), surpassing previous studies. We further investigated the role of different EEG frequency bands—Alpha, Beta, Gamma, and High Gamma—in imagined speech decoding. The results indicate that the Beta and Alpha bands are most effective for decoding, offering reliable neural signal representations. By improving the capacity to classify imagined speech and identifying the most effective EEG frequency bands for decoding, these findings lay the groundwork for more refined and accessible BCI applications in the future.

THINKING OUT LOUD: UNVEILING BRAIN OSCILLATIONS IN VOWEL AND SEMANTIC DECODING.

VALES CORTINA, IBON
2023/2024

Abstract

This study explores the neural mechanisms underlying imagined speech and its potential for brain-computer interface (BCI) applications, particularly for individuals with severe communication impairments. Using EEG and a novel deep learning model, we aimed to decode linguistic components from imagined speech, focusing on vowels and semantic categories. Our model, which integrates convolutional and recurrent neural network architectures, was validated through the classification of imagined speech, overt speech, and silence, achieving a significant accuracy of 81.76%. High accuracy was also obtained in vowel classification during imagined speech (92.06%) and in semantic categorization (84.88%), surpassing previous studies. We further investigated the role of different EEG frequency bands—Alpha, Beta, Gamma, and High Gamma—in imagined speech decoding. The results indicate that the Beta and Alpha bands are most effective for decoding, offering reliable neural signal representations. By improving the capacity to classify imagined speech and identifying the most effective EEG frequency bands for decoding, these findings lay the groundwork for more refined and accessible BCI applications in the future.
2023
THINKING OUT LOUD: UNVEILING BRAIN OSCILLATIONS IN VOWEL AND SEMANTIC DECODING. Abstract: This study explores the neural mechanisms underlying imagined speech and its potential for brain-computer interface (BCI) applications, particularly for individuals with severe communication impairments. Using EEG and a novel deep learning model, we aimed to decode linguistic components from imagined speech, focusing on vowels and semantic categories. Our model, which integrates convolutional and recurrent neural network architectures, was validated through the classification of imagined speech, overt speech, and silence, achieving a significant accuracy of 81.76%. High accuracy was also obtained in vowel classification during imagined speech (92.06%) and in semantic categorization (84.88%), surpassing previous studies. We further investigated the role of different EEG frequency bands—Alpha, Beta, Gamma, and High Gamma—in imagined speech decoding. The results indicate that the Beta and Alpha bands are most effective for decoding, offering reliable neural signal representations. By improving the capacity to classify imagined speech and identifying the most effective EEG frequency bands for decoding, these findings lay the groundwork for more refined and accessible BCI applications in the future.
Imagined Speech
BCI
EEG
DL Classification
Frequency Bands
File in questo prodotto:
File Dimensione Formato  
Entrega UNIPD.pdf

accesso aperto

Dimensione 2.02 MB
Formato Adobe PDF
2.02 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/73281