EEG-based Syllable-Level Voice Activity Detection.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) offer a direct communication pathway between the brain and external devices.
• Electroencephalogram (EEG) signals are increasingly utilized for decoding neural activity related to speech.
• Accurate voice activity detection (VAD) is crucial for developing effective speech BCIs.

Purpose of the Study:

• To investigate syllable-level voice activity detection (VAD) using electroencephalogram (EEG) signals.
• To develop and evaluate an EEG-based VAD model for distinguishing speech-related brain activity.
• To explore the potential of EEG-based VAD in future self-paced speech BCI applications.

Main Methods:

• Collected EEG data from 10 participants during auditory (listening) and speech (pronouncing syllables) tasks.
• Applied Speech-Based VAD to label auditory stimuli and voice recordings, creating corresponding brain activity labels.
• Classified EEG states into resting and active (listening or pronouncing) using the developed VAD model.

Main Results:

• The EEG-based VAD model achieved accuracies of 90.93% for speech production and 69.57% for auditory speech tasks.
• Cross-subject classification accuracies were 72.63% and 61.15% for the respective tasks, indicating lower performance.
• No significant correlation was found between time window length and classification accuracy.

Conclusions:

• EEG-based VAD shows promise for applications in speech decoding and self-paced BCIs.
• The developed model demonstrates feasibility for identifying speech-related EEG activity.
• Further research is needed to improve cross-subject generalization for robust BCI performance.