Low-Latency Auditory Spatial Attention Detection Based on Spectro-Spatial Features from EEG.

Area of Science:

• Neuroscience
• Signal Processing
• Brain-Computer Interfaces

Background:

• Auditory attention detection using brain signals is crucial for applications like speech processing and understanding the cocktail party effect.
• Existing methods often rely on correlating brain signals with auditory stimuli.
• Alpha band (8-13 Hz) EEG signals have shown potential for localizing auditory stimuli.

Purpose of the Study:

• To develop a method for detecting auditory spatial attention (left/right) without requiring auditory stimuli as references.
• To investigate the use of spectro-spatial features derived from alpha band power for attention detection.

Main Methods:

• A novel spectro-spatial feature extraction technique was proposed.
• The technique leverages the topographic specificity of alpha band power in electroencephalography (EEG) signals.
• Auditory spatial attention was detected based on these extracted features.

Main Results:

• The proposed neural approach achieved 81.7% accuracy with a 1-second decision window.
• Accuracy increased to 94.6% with a 10-second decision window.
• The method significantly outperformed other competitive models in all test cases.

Conclusions:

• It is feasible to detect auditory spatial attention using EEG signals without auditory stimuli.
• The proposed spectro-spatial feature extraction technique is effective and robust.
• This approach offers a promising advancement for brain-computer interface applications requiring attention detection.