XAGnet: Cross-Attention Graph Network for Detecting Auditory Attention in Ear-EEG Signals.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Auditory Attention Detection (AAD) is crucial for advanced brain-computer interfaces and hearing technologies.
• Complex auditory environments pose significant challenges for current AAD systems.

Purpose of the Study:

• To propose XAGnet, a novel method for AAD using ear-centered EEG (ear-EEG).
• To model intra-ear and inter-ear neural dependencies for improved attention detection.

Main Methods:

• Utilized Graph Convolutional Networks (GCNs) on left and right ear-EEG for intra-ear feature extraction.
• Implemented a cross-attention mechanism to capture inter-ear neural interactions.
• Employed multi-class classification to identify attended speaker locations.

Main Results:

• XAGnet demonstrated superior performance compared to baseline models on a public ear-EEG dataset.
• The method successfully detected auditory attention among four distinct speakers.
• Effectiveness of modeling both intra-ear and inter-ear dependencies was highlighted.

Conclusions:

• XAGnet offers a promising advancement in Auditory Attention Detection.
• Leveraging ear-EEG with sophisticated neural network architectures can enhance brain-computer interface capabilities.
• The proposed method shows potential for real-world applications in noisy auditory settings.