[Paper Review] Voice of Your Brain: Cognitive Representations of Imagined Speech,Overt Speech, and Speech Perception Based on EEG

Seo‐Hyun Lee, Young Eun Lee+1 more|arXiv (Cornell University)|May 31, 2021

EEG and Brain-Computer Interfaces21 references4 citations

TL;DR

This study demonstrates that imagined speech and overt speech produce distinct, individualized EEG patterns across nine subjects, enabling reliable speaker identification using a single-channel EEG and a deep neural network that captures temporal-spectral-spatial features. The results show that imagined speech, in particular, exhibits strong subject-specific neural signatures, especially at the T7 electrode, supporting its practical use in secure, single-channel brain-computer interfaces for personal identification.

ABSTRACT

Every people has their own voice, likewise, brain signals dis-play distinct neural representations for each individual. Al-though recent studies have revealed the robustness of speech-related paradigms for efficient brain-computer interface, the dis-tinction on their cognitive representations with practical usabil-ity still remains to be discovered. Herein, we investigate the dis-tinct brain patterns from electroencephalography (EEG) duringimagined speech, overt speech, and speech perception in termsof subject variations with its practical use of speaker identifica-tion from single channel EEG. We performed classification ofnine subjects using deep neural network that captures temporal-spectral-spatial features from EEG of imagined speech, overtspeech, and speech perception. Furthermore, we demonstratedthe underlying neural features of individual subjects while per-forming imagined speech by comparing the functional connec-tivity and the EEG envelope features. Our results demonstratethe possibility of subject identification from single channel EEGof imagined speech and overt speech. Also, the comparison ofthe three speech-related paradigms will provide valuable infor-mation for the practical use of speech-related brain signals inthe further studies.

Motivation & Objective

• To investigate whether EEG signals during imagined speech, overt speech, and speech perception contain subject-specific neural representations.
• To evaluate the feasibility of speaker identification using single-channel EEG in practical BCI applications.
• To compare the cognitive neural patterns—functional connectivity and EEG envelope—across imagined speech, overt speech, and speech perception to identify individual differences.
• To determine the optimal EEG channel for subject identification, focusing on practicality and performance.

Proposed method

• EEG data were collected from nine subjects across four conditions: imagined speech, overt speech, speech perception, and resting state, with 300 trials per condition.
• Signals were preprocessed using a 5th-order Butterworth filter (30–120 Hz) and baseline correction over 500 ms before trial onset.
• A deep neural network architecture was designed to extract temporal, spectral, and spatial features from single-channel EEG, enabling end-to-end classification of speaker identity.
• Functional connectivity was analyzed using Phase-Lag Index (PLV) in Broca’s and Wernicke’s areas to compare neural network dynamics during imagined speech versus resting state.
• EEG envelope features were computed and averaged across subjects to identify temporal patterns unique to each individual during imagined speech.
• Classification performance was evaluated using accuracy, with t-SNE visualization to assess inter-class separability of neural representations.

Experimental results

Research questions

• RQ1 Can individual speakers be reliably identified from single-channel EEG during imagined speech, overt speech, and speech perception?
• RQ2 How do the neural representations of imagined speech compare to those of overt speech and speech perception in terms of subject-specificity?
• RQ3 Which EEG channel provides the highest performance for speaker identification in a single-channel setup?
• RQ4 Do functional connectivity patterns in speech-related brain regions (Broca’s and Wernicke’s areas) differ significantly between subjects during imagined speech?
• RQ5 To what extent do EEG envelope dynamics during imagined speech reflect individual speech characteristics?

Key findings

• The T7 electrode showed significantly superior speaker identification accuracy (p < 0.001, χ² = 32.51) compared to other channels, indicating its strong potential for practical ear-EEG devices.
• Imagined speech and overt speech produced more distinct, subject-specific neural patterns than speech perception or resting state, as confirmed by t-SNE visualization and classification accuracy.
• Subject 3 exhibited the most distinct neural signature in both functional connectivity and EEG envelope, correlating with the highest classification performance.
• Functional connectivity in Broca’s area was decreased during imagined speech for most subjects, but increased for subject 3, suggesting individual differences in neural activation patterns.
• The EEG envelope of each subject showed unique peak patterns between 500–1500 ms, with subject 3 displaying a radiated feature pattern that aligned with high discriminability.
• Speech perception showed slightly higher identification accuracy (56.26 ± 1.87%) than resting state (53.87 ± 3.49%), though performance was limited by single-session data availability.