Auditory Saliency Classification Using EEG Data

Abstract

We investigate if starting from EEG signals it is possible to classify salient audio events. We set up an experiment and collected EEG data with proper audio stimuli. We trained several kNN classifiers. As the EEG signals under different frequency bands are related to different brain activities, we analyzed EEG data from each single electrode and for each single frequency band, as well as EEG data from combined electrodes and frequency bands. From our preliminary analysis we found that for each frequency band there is a set of electrodes that permits to achieve better classification results and these electrodes are related to specific brain regions. We propose a cross-subject approach, where different classifiers are trained on each single individual, and tested on data collected from all the others. This initial investigation suggests several hints to improve this classification task.