The performance of EEG-based auditory attention decoding according to the speech volume in a dichotic listening task: a preliminary study

Abstract

Auditory attention decoding (AAD) has been developed to detect the attended speech using electroencephalography. Recently, a real-time AAD method using linear decoder model has been introduced lead to expand the application of AAD in terms of auditory brain computer interface. However, people focus on sounds of varying volumes in everyday conversation, so the question of whether AAD is possible in various sound stimuli arises. The present study aimed to investigate the effect of the difference in volume between two competing speech in a dichotic listening paradigm for online AAD. Most comfortable level (MCL) and dichotic speech recognition threshold (SRT) were evaluated. And the 'Sound level for Speech intelligibility (SI) of 90%' and 'Sound Level for SI 50%' were also calculated. In online AAD task, four different sound level (MCL, MCL-20dBA, Sound level for SI of 90% and SI 50%) were introduced as attended speech, while the sound level of ignored speeches was fixed at the MCL. There was no difference in AAD performance based on the four sound levels conditions. In addition, volume difference was not significantly correlated with the individual decoder accuracy. This preliminary study identified that lowering attended speech volume to SRT level had no effect on AAD.