Classification of auditory attention focuses during speech perception

Area of Science:

• Neuroscience
• Cognitive Science
• Signal Processing

Background:

• Passive brain-computer interfaces (BCIs) decode cognitive and emotional states using neurophysiological signals.
• Monitoring attentional state is crucial for passive BCIs.
• Existing research primarily classifies attention levels (high vs. low), with limited focus on attention direction during speech perception.

Purpose of the Study:

• To develop and evaluate a method for recognizing attention focuses during auditory speech perception using electroencephalography (EEG).
• To classify whether a subject is attending to a call sign or a number within a spoken sentence.

Main Methods:

• Utilized electroencephalography (EEG) data from 15 subjects.
• Subjects were instructed to focus on either a call sign or a number during listening tasks.
• Proposed a novel algorithm combining Common Spatial Pattern (CSP), Short-Time Fourier Transformation (STFT), and Discriminative Canonical Pattern Matching (DCPM) for EEG pattern classification.

Main Results:

• Achieved an average classification accuracy of 78.38% for single-trial classification.
• Reached a peak accuracy of 93.93% in classifying attention focuses.
• Demonstrated the effectiveness of the proposed CSP-STFT-DCPM algorithm.

Conclusions:

• The developed algorithm is effective in classifying auditory attention focuses during speech perception.
• This research contributes to advancing passive BCIs by enabling the monitoring of specific attention targets.
• The findings have implications for understanding and potentially enhancing human-computer interaction in complex auditory environments.