Auditory attention detection in cocktail-party: A microstate study

Abstract

Healthy human brain can easily attend to one auditory stimulus and filter out the other stimuli, described in the terms of “selective auditory attention” or referred to as the cocktail party phenomenon. This paper proposes a new real-time mapping-based selective auditory attention detection (SAAD) system using only EEG signals. In the proposed system, several quasi-stable maps, called EEG microstates, are extracted. After the optimization of the microstates topographies, the temporal dynamics of the microstates are used for the classification of auditory attention direction. The evaluation results obtained in different listening conditions (i.e., dichotic and spatial hearing) show that the use of an integrated set of 6 microstate topographies best describes the data and provides better classification performance. Due to the nature of EEG microstates in reflecting the fundamentals of brain information processing, this study provides an appropriate way of reaching the underlying mechanism of selective attention. The proposed system based on microstates can effectively detect the attention direction in dynamic scenarios, where the listener’s attention might be switched in seconds, making the model suitable for real-time applications. Furthermore, the proposed auditory attention detection system is advantageous in the sense that the detection of attention is performed without using speech signals.