Kosmyna, N., Balyan, A., Hauptmann, E. (2023). Target Speaker Detection with EEG/EOG Glasses: A Pilot Study. In: Arai, K. (eds) Proceedings of the Future Technologies Conference (FTC) 2022, Volume 2. FTC 2022 2022. Lecture Notes in Networks and Systems, vol 560. Springer, Cham. https://doi.org/10.1007/978-3-031-18458-1_30

Abstract

 Recent studies have examined the use of electroencephalography (EEG) and electrooculography (EOG) for target speaker detection in multi-speaker paradigms. The use of small, mobile and low-channel EEG and EOG-based devices is a recent trend due to the user-friendly nature of such solutions and their deployment outside the lab. The current study is focused on exploring if a low-channel EEG/EOG device can be used for target speaker detection. The performance of a low-channel EEG/EOG device was compared with a research-grade EEG headset for target speaker identification in a dichotic listening paradigm. 14 users took part in a study where a cocktail party problem was simulated and two audio streams were played simultaneously. Participants were asked to attend to one audio stream at a time. Audio exempts belonging to various genres of stories were simultaneously presented via the left and right audio streams as stimuli. Convolutional Neural Networks (CNNs) were used for classification of the target speaker in a dichotic listening paradigm. Low-channel EEG/EOG device identified the target speakers with an average accuracy of 74% and the high-channel EEG system achieved an average accuracy of 79%. The results provide encouraging results to continue exploring the feasibility of using wearable, low-channel EEG/EOG devices for auditory attention tasks.