Event-based Auditory Attention Decoding Dataset Aarhus University

Description

Methods

This dataset contains EEG (Scalp + in-ear) recordings from 24 normal hearing and native Danish-speaking subjects, who participated sequentially in the study with four paradigms. All recordings were conducted in an acoustically shielded listening room with 0.4 s of reverberation time, using 32 scalp electrodes and a left and right ear-EEG earpiece with six electrodes on each earpiece. Details of four paradigms are as follows:

1. Paradigm 1 - word category oddball comprises 16 trials. In each trial, the subject was presented with a sequence of two different classes of spoken words: animal names and cardinal numbers, or color names and cardinal numbers from a loudspeaker. The subject was asked to pay attention to the target events, which were animal names or color names, and passively count them.
2. Paradigm 2 - word category with competing speakers comprises 20 trials and uses similar sequences of discrete spoken words as in Paradigm 1. However, in this paradigm, two competing streams were presented simultaneously from two loudspeakers placed 60 degrees to the left and right. The subject was asked to pay attention to only the target events in one of the streams and count them while disregarding the other stream.
3. Paradigm 3 - competing speech streams with targets comprises 20 trials and is similar to Paradigm 2. In each trial, the subject was presented with two competing streams of different continuous stories. In each stream, one class of words was predefined as target words, including animal names, human names, color names, and plant species. The subject was asked to attend to one of the two streams (left or right) and focus on the target words of that stream. At the end of the trial, the subject answered a question about the target words and received feedback.
4. Paradigm 4 - competing speech streams without targets was designed to simulate a real-world scenario of selective listening in a setting with multiple sound sources. Two competing streams with two different stories from two loudspeakers were presented to the subject. The subject was instructed to attend to one stream while disregarding the other in each trial. Following each trial, the subject was probed with a question about the content and was provided with feedback.

For details of the dataset and its structure, please refer to the README file.