Performance Analysis of Stimulus Reconstruction and Spatial Lateralization for EEG-Based AAD Using DL Applications

This study looks into listener attention in a more realistic four-talker setting to make up for the shortcomings of previous Auditory Attention Decoding (AAD) research, which mostly looked at simplified two-talker situations. A new AAD database was created that records the electroencephalogram (EEG) responses of listeners while they listen to four different dialogues at the same time in different locations. We used cortical spatial lateralization and stimulus reconstruction (SR)-based AAD methods to figure out what people were paying attention to. The results show that SR works very well, with a decoding accuracy of 77.5% over a 60-second period, which is much higher than the 25% chance level. Also, Auditory Spatial Attention Detection (ASAD) methods did an amazing job of generalizing, getting 94.7% accuracy with DenseNet-3D in just one second. These results show that it is possible to accurately decode auditory attention in very complicated sound environments. This opens the door to more advanced real-world uses in brain-computer interfaces and assistive listening devices.