Contribution
Classifying the Direction of Auditory Attention from EEG Signals Using Deep Learning with a Short Decision Window
* Presenting author
Abstract:
Auditory attention decoding (AAD) from electroencephalography (EEG) focuses on identifying the direction of attention in noisy environments. The primary objective is to determine the most effective methods for isolating the attended auditory source, mainly when two audio streams are present concurrently. Accurately tracking the direction of auditory attention over short time frames is crucial, especially considering that previous research has shown that linear regression techniques yield better results with longer decision windows.Recent advancements in deep learning, however, have demonstrated that neural networks can successfully detect attended locations even within shorter epochs, such as 1-second segments. This study introduces a deep learning model that incorporates an attention mechanism to decode auditory attention from EEG signals while participants engage in tasks involving concurrent auditory inputs. We segment EEG recordings into short decision epochs, classifying the direction of attention as either left or right. To facilitate thorough evaluation, we visualize the predicted attention direction, enabling tracking of attentional shifts and paving the groundwork for potential real-time applications.Our findings demonstrate that the proposed deep learning approach outperforms traditional linear methods in terms of classification performance. This research underscores the effectiveness of nonlinear techniques in AAD.