Auditory-GAN: deep learning framework for improved auditory spatial attention detection.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Auditory attention detection using electroencephalography (EEG) faces challenges with limited online data and low-latency detection.
• Existing methods struggle with data scarcity and real-time processing requirements.

Purpose of the Study:

• To introduce Auditory-GAN, a deep generative adversarial network auxiliary, for enhanced EEG data generation and auditory spatial detection.
• To address the limitations of data scarcity and low latency in auditory attention detection.

Main Methods:

• A spectro-spatial feature extraction (SSF) module was developed to capture topographic specificity of alpha power from EEG signals.
• An auditory generative adversarial network auxiliary (AD-GAN) classifier was designed to synthesize augmented EEG data, addressing data scarcity.
• The system integrates SSF for feature extraction and AD-GAN for classification and data augmentation.

Main Results:

• Auditory-GAN successfully generates convincing EEG data, validated on the KUL dataset.
• Achieved a high spatial attention detection accuracy of 98.5% with a 10-s decision window using 64-channel EEG data.
• Outperformed state-of-the-art models across various channel counts (64 to 32) in comparative analyses.

Conclusions:

• Auditory-GAN offers a robust solution for auditory spatial attention detection, effectively overcoming data limitations.
• The proposed deep learning approach demonstrates significant improvements in accuracy and efficiency for EEG-based attention detection.
• The model's code is publicly available, facilitating further research and application.