Detecting Attention Shift from Neural Response Based on Beat-frequency-modulated Musical Excerpts

Abstract

This paper presents a new approach for detecting attention from auditory steady-state responses (ASSR) by using musical excerpts. The feature extraction process for electroencephalogram (EEG) signal is combined with a support vector machine as a binary discriminator. A novel modulation that emphasizes the beat timing of the excerpts has enhances the EEG response. Thanks to the beat-locked epoch extraction and additional information that signals the locked excerpt, the estimation errors are less than 8% using only ten seconds of data. Contrary to our expectations, a waveform-averaging method outperforms a harmonic filter bank and a bin-subtraction methods in the frequency domain for feature extraction. Overall, attention estimation from EEGs using musical excerpts as stimuli has been successfully achieved, which represents significant progress towards the development of a mass-EEG measurement system.