Classification of Three Intentions Based on Auditory Steady State Response to Natural Sound

Abstract

Brain-computer interface (BCI) has been proposed as a means of communication for people with severe physical disabilities. BCI based on auditory steady-state response (ASSR) in evoked electroencephalogram (ASSR-BCI) has an advantage of low-risk of auditory stimuli and does not need lengthy training to elicit ASSR by amplitude-modulated acoustic stimuli. Recent work reported that auditory stimulation with amplitude modulation of natural sounds improved the classification accuracy of ASSR-BCI. However, most of the conventional ASSR-BCIs including the aforementioned one have been limited to a two-class system which has only two options for user’s intentions at a time. The present study proposed a three-class ASSR-BCI by combining amplitude-modulated natural sounds of a trickling brook and birdsong for amplification of ASSR with a mental task for class expansion. The amplitude of ASSR to the amplitude-modulated natural sound stimuli became larger than that to a conventional amplitude-modulated tone for more subjects. Moreover, the proposed three-class ASSR-BCI showed the classification accuracy exceeded 60 % by using optimal auditory stimuli depending on subjects which suggests the impact of mental task. A significant difference from a chance level found in the classification accuracy evidenced the feasibility of the proposed three-class ASSR-BCI.