Find research datasets worth reusing
Search datasets from major research repositories and use ShareScore to quickly assess how well each record supports discovery, access, and reuse.
5
datasets available to search
ShareScore release 0.9.0
Dataset results
5 results for “Auditory Attention Decoding”
MAD-EEG: an EEG dataset for decoding auditory attention to a target instrument in polyphonic music
<p>The <em><strong>MAD-EEG Dataset</strong></em> is a research corpus for studying EEG-based auditory attention decoding to a target instrument in polyphonic music. </p> <p>The dataset consists of 20-channel EEG responses to music recorded from 8 subjects while attending to a particular instrument in a music mixture. </p> <p>For further details, please refer to the paper: <em><a href="https://hal.archives-ouvertes.fr/hal-02291882/document">MAD-EEG: an EEG dataset for decoding auditory attention to a target instrument in polyphonic music</a>.</em></p> <p>If you use the data in your research, please reference the paper (not just the Zenodo record):</p> <pre><code>@inproceedings{Cantisani2019, author={Giorgia Cantisani and Gabriel Trégoat and Slim Essid and Gaël Richard}, title={{MAD-EEG: an EEG dataset for decoding auditory attention to a target instrument in polyphonic music}}, year=2019, booktitle={Proc. SMM19, Workshop on Speech, Music and Mind 2019}, pages={51--55}, doi={10.21437/SMM.2019-11}, url={http://dx.doi.org/10.21437/SMM.2019-11} }</code></pre> <p> </p>
Audiovisual, Gaze-controlled Auditory Attention Decoding Dataset KU Leuven (AV-GC-AAD)
<p>This dataset is described in detail in the following journal paper [1]:<br>Rotaru, I., Geirnaert, S., Heintz, N., Van de Ryck, I., Bertrand, A., & Francart, T. (2024). What are we really decoding? Unveiling biases in EEG-based decoding of the spatial focus of auditory attention. Journal of Neural Engineering, 21(1), 016017.<br><a href="https://iopscience.iop.org/article/10.1088/1741-2552/ad2214/meta">https://iopscience.iop.org/article/10.1088/1741-2552/ad2214/meta</a></p> <p><em><strong> If using this dataset, please cite the original paper above and the current Zenodo repository. </strong></em></p> <p><strong>Note from the authors: </strong>Recent evaluations reveal that various published AAD (Auditory Attention Decoding) algorithms do not achieve significant above-chance performance on this AV-GC-AAD dataset, and in particular on the two gaze-incongruent conditions 'MovingVideo' and 'MovingTargetNoise'). This suggests that previously reported successes may have been largely influenced by eye gaze confounds present in other datasets, which can be exploited as shortcuts by machine learning algorithms. Despite these findings, poor performance on the AV-GC-AAD dataset is often dismissed, with reasons cited such as insufficient training data, high heterogeneity in audiovisual conditions, or the claim that participants were unable to focus their auditory attention due to the complexity of the instructions.</p> <p>To address these concerns, we provide a supplementary technical report (and accompanying code), showcasing results from a simple linear stimulus reconstruction AAD algorithm applied to this dataset. Our findings demonstrate that high AAD accuracy can be achieved within individual conditions, and that the model generalizes across conditions, new subjects, and even across different datasets.</p> <p><a title="https://doi.org/10.48550/arxiv.2412.01401" href="https://doi.org/10.48550/arXiv.2412.01401" target="_blank" rel="noreferrer noopener">Report</a> | <a title="https://github.com/alexanderbertrandlab/linear-stimulus-reconstruction-aad-av-gc-aad-dataset" href="https://github.com/AlexanderBertrandLab/linear-stimulus-reconstruction-AAD-AV-GC-AAD-dataset" target="_blank" rel="noreferrer noopener">Matlab Code</a></p> <p>Through this report, we aim to remove any doubts that the AV-GC-AAD dataset's limitations are the primary cause of AAD algorithms failing to exceed chance-level performance. Additionally, this report and its accompanying code offer a simple baseline evaluation procedure, which can serve as a minimal benchmark for testing more advanced AAD algorithms on this dataset.</p> <p><em>When reporting results on this data set, it is good practice to show performance for each condition separately, since 2 of the 4 conditions still contain gaze shortcuts, which could be exploited by machine learning algorithms. </em></p> <p>________________________________________________________________________________</p> <p><strong>Dataset description</strong></p> <p>This work was performed at ExpORL, Dept. Neurosciences, KU Leuven and Dept. Electrical Engineering (ESAT), KU Leuven (Belgium), with the goal of investigating and controlling for the effect of gaze during a competing listening task.</p> <p>The full dataset contains EEG and EOG data collected from 16 normal-hearing subjects, during a competing listening task, where the subjects were instructed to focus on one of two competing speech signals. However, subjects 2, 5 and 6 were excluded from the online repository due to not consenting to sharing their data in a public database (cf. signed informed consents approved by KU Leuven Ethical Committee). EEG recordings were conducted in a soundproof, electromagnetically shielded room at ExpORL, KU Leuven. The BioSemi ActiveTwo system was used to record 64-channel EEG signals at 8196 Hz sample rate. Additionally, the participants' gaze movements were measured via 4 EOG (electrooculography) electrodes placed symmetrically around the eyes. </p> <p>The audio signals were administered to each subject at 65 dB SPL through a pair of insert phones (Etymotic ER10). In some experimental trials, the video depicting the attended talker was also presented on the screen. The original presented speech and video stimuli (.wav and .mp4 files) are excluded from the dataset due to copyrights. However, the acoustic envelopes of the attended and unattended audio stimuli are calculated and included in the dataset (see below). <br>The experiments were conducted using custom-made Python scripts.</p> <p>The experimental trials were split into 2 blocks. Each block consisted of the following sequence of conditions: MovingVideo, MovingTargetNoise, NoVisuals, StaticVideo. The auditory task was the same for all conditions: the subjects had to attend to one of the two presented talkers, as indicated by an arrow on the screen. The visual task differed across conditions:</p> <ul> <li>MovingVideo: the subjects had to follow the moving video of the to-be-attended speaker presented on a randomized horizontal trajectory on the screen.</li> <li>MovingTargetNoise: the subjects had to follow a moving cross-hair presented on a randomized horizontal trajectory on the screen.</li> <li>NoVisuals: a black screen was presented and the subjects had to fixate on an imaginary point in the center of the screen while minimizing the eye movements.</li> <li>StaticVideo: the subjects had to fixate the static video of the to-be-attended speaker presented on the same side with the audio stimulus of the attended speaker.</li> </ul> <p>The full description of all experimental conditions can be consulted in [1].</p> <p>Each trial/condition lasted for 10 minutes, with a <strong><em>spatial switch</em></strong> in attention after 5 minutes (i.e., the presented speech stimuli were programmed to swap sides - from L to R or vice versa, such that after the switch the subjects kept listening to the same speaker, but coming from the opposite spatial location). This means that the participant kept attending to the same speaker throughout an entire trial. To keep the subjects motivated, they had to answer one comprehension question related to the attended acoustic stimulus after each trial.</p> <p>For each subject, there is a<strong> .mat file</strong> containing the following variables:<br><strong>conditionID:</strong> the condition ID for each trial <br><strong>data</strong>: the preprocessed EEG and EOG data for each trial (first 64 channels are EEG, last 4 are EOG)<br><strong>fs:</strong> the sampling rate of the EEG, EOG and stimuli envelopes<br><strong>initAttention</strong>: the initial spatial location of the attended stimulus for each trial<br><strong>metadata</strong>: the original metadata (e.g. channel names, triggers) saved in the raw .bdf files for each trial<br><strong>params</strong>: the filtering parameters used for each trial<br><strong>randomization</strong>: the randomization parameters (e.g. presented stimuli, attention switch times etc.) for each trial<br><strong>stimulus</strong>: the precalculated envelopes for the attended and unattended stimuli for each trial<br><strong>subjID</strong>: the anonymised ID of the current subject</p> <p><strong>Preprocessing EEG and EOG</strong></p> <p>All the following preprocessing steps were applied per trial. The EEG was initially downsampled using an antialiasing filter from 8192 Hz to 256 Hz. The data was then filtered between 1–40 Hz using a zero-phase Chebyshev filter (type II, with 80 dB attenuation at 10% outside the passband). Finally, downsampling to 128 Hz was performed to speed up computation.</p> <p><strong>Speech envelopes extraction</strong></p> <p>The original speech signals at 44100 Hz were downsampled to 8192 Hz (to match the EEG sampling rate). They were then passed through a gammatone filterbank, which roughly approximates the spectral decomposition as performed by the human auditory system. Per subband, the audio envelopes were extracted, and their dynamic range was compressed using a power-law operation with exponent 0.6 (as proposed in [2]). Each subband was then bandpass-filtered with the same filter used for the EEG data. The resulting subband envelopes were then summed to construct a single broadband envelope. Finally, the envelope signals were downsampled to 128 Hz to match the sampling rate of the preprocessed EEG.</p> <p><strong>Notes</strong></p> <ol> <li>For subjects 1-3, 6 trials corresponding to 3 conditions (MovingVideo, NoVisuals, StaticVideo) were measured.</li> <li>For subjects 4-16, 8 trials corresponding to 4 conditions (MovingVideo, MovingTargetNoise, NoVisuals, StaticVideo) were measured.</li> <li>For subject 14, trial 2 from the StaticVideo condition was not recorded due to some technical problems.</li> <li>In the dataset, 'FixedVideo' is the alias name for the 'StaticVideo' condition described in [1].</li> <li>The EEG/EOG data was not referenced. Before further analysis, rereferencing the data (e.g., to an arbitrary EEG channel, or the common-average of all channels) is necessary to achieve a better common-mode rejection and thus increase the SNR of recorded data. (for details, see https://www.biosemi.com/faq/cms&drl.htm)</li> </ol> <p><strong>References</strong></p> <p>[1] Rotaru, Iustina, et al. "What are we really decoding? Unveiling biases in EEG-based decoding of the spatial focus of auditory attention." <em>Journal of Neural Engineering</em> 21.1 (2024): 016017.</p> <p>[2] Biesmans, Wouter, et al. "Auditory-inspired speech envelope extraction methods for improved EEG-based auditory attention detection in a cocktail party scenario." <em>IEEE Transactions on neural systems and rehabilitation engineering</em> 25.5 (2016): 402-412.</p>
EEG and audio dataset for auditory attention decoding
<p>This dataset contains EEG recordings from 18 subjects listening to one of two competing speech audio streams. Continuous speech in trials of ~50 sec. was presented to normal hearing listeners in simulated rooms with different degrees of reverberation. Subjects were asked to attend one of two spatially separated speakers (one male, one female) and ignore the other. Repeated trials with presentation of a single talker were also recorded. The data were recorded in a double-walled soundproof booth at the Technical University of Denmark (DTU) using a 64-channel Biosemi system and digitized at a sampling rate of 512 Hz. Full details can be found in:</p> <ul> <li><strong>Søren A. Fuglsang, Torsten Dau & Jens Hjortkjær (2017): Noise-robust cortical tracking of attended speech in real-life environments. <em>NeuroImage</em>, 156, 435-444</strong></li> </ul> <p>and</p> <ul> <li><strong>Daniel D.E. Wong, Søren A. Fuglsang, Jens Hjortkjær, Enea Ceolini, Malcolm Slaney & Alain de Cheveigné: A Comparison of Temporal Response Function Estimation Methods for Auditory Attention Decoding. Frontiers in Neuroscience, </strong><a href="https://doi.org/10.3389/fnins.2018.00531">https://doi.org/10.3389/fnins.2018.00531</a></li> </ul> <p>The data is organized in format of the publicly available <a href="https://zenodo.org/record/1198430">COCOHA Matlab Toolbox</a>. The preproc_script.m demonstrates how to import and align the EEG and audio data. The script also demonstrates some EEG preprocessing steps as used the Wong et al. paper above. The AUDIO.zip contains wav-files with the speech audio used in the experiment. The EEG.zip contains MAT-files with the EEG/EOG data for each subject. The EEG/EOG data are found in <strong>data.eeg</strong> with the following channels:</p> <ul> <li>channels 1-64: scalp EEG electrodes</li> <li>channel 65: right mastoid electrode</li> <li>channel 66: left mastoid electrode</li> <li>channel 67: vertical EOG below right eye</li> <li>channel 68: horizontal EOG right eye</li> <li>channel 69: vertical EOG above right eye</li> <li>channel 70: vertical EOG below left eye</li> <li>channel 71: horizontal EOG left eye</li> <li>channel 72: vertical EOG above left eye</li> </ul> <p>The <strong>expinfo</strong> table contains information about experimental conditions, including what what speaker the listener was attending to in different trials. The expinfo table contains the following information:</p> <ul> <li>attend_mf: attended speaker (1=male, 2=female)</li> <li>attend_lr: spatial position of the attended speaker (1=left, 2=right)</li> <li>acoustic_condition: type of acoustic room (1= anechoic, 2= mild reverberation, 3= high reverberation, see Fuglsang et al. for details)</li> <li>n_speakers: number of speakers presented (1 or 2)</li> <li>wavfile_male: name of presented audio wav-file for the male speaker</li> <li>wavfile_female: name of presented audio wav-file for the female speaker (if any)</li> <li>trigger: trigger event value for each trial also found in data.event.eeg.value</li> </ul> <p>DATA_preproc.zip contains the preprocessed EEG and audio data as output from preproc_script.m.</p> <p>The dataset was created within the <a href="https://cocoha.org/">COCOHA</a><a href="https://cocoha.org/"> Project</a>: Cognitive Control of a Hearing Aid</p>
EEG Dataset for 'Decoding of selective attention to continuous speech from the human auditory brainstem response' and 'Neural Speech Tracking in the Theta and in the Delta Frequency Band Differentially Encode Clarity and Comprehension of Speech in Noise'.
<p>The repository contains the unprocessed EEG data recorded for the publications [1, 2]. For convenience, the onsets of the EEG data provided here are time-aligned with the onsets of the audio books in the 'audiobooks' folder, and the EEG data are provided in HDF5 format. Please refer to the original version of this dataset for more details.</p> <p>More details, as well as the original data files, are available at the original repository <a href="https://doi.org/10.5281/zenodo.7086209">here</a>.</p> <p>Examples of using these data (preprocessing, fitting linear models) can be found <a href="https://github.com/Mike-boop/trf-examples">here</a>.</p> <p>The English conditions (clean, lb, mb, hb, fM, fW) comprised a single recording session. The Dutch conditions (cleanDutch, lbDutch, mbDutch, hbDutch) comprised a separate recording session. You see which participants took part in each session in session_info.json.</p> <p>Please note some details about the stimulus presentation for the various listening conditions:</p> <ul> <li>English speech-in-babble-noise (lb, mb, hb): babble noise was played by itself for one second before the audiobook track began. The babble noise was also played for one second after the audiobook track ended. Therefore, you should discard the first second and the last second from these trial during your analysis.</li> <li>Dutch speech-in-babble-noise (lbDutch, mbDutch, hbDutch): the story (narrated in Dutch) was played by itself for one second before the babble noise track began. Then, the babble noise was increased linearly in amplitude for one second. Therefore, you should discard the first two seconds from these trials during your analysis.</li> <li>Dutch in quiet, and Dutch-in-babble-noise (cleanDutch, lbDutch, mbDutch, hbDutch): some English sentences were embedded in the Dutch narratives in order to encourage attention. You should crop these from your analysis. The onsets and offsets of the English sentences (in samples, at 44100Hz) are provided in the audiobooks/*Dutch/english_onsets_info.json files.</li> <li>Competing-speakers conditions (fM, fW): sometimes the attended track is longer than the unattended track, or vice-versa. The onsets of both tracks are aligned. You should crop the trial to the length of the shortest track for your analysis.</li> </ul> <p>If you use this data, please cite the original publications, as well as this repository [1,2,3].</p> <p>[1] Etard O, Kegler M, Braiman C, Forte A E and Reichenbach T. “Decoding of selective attention to continuous speech from the human auditory brainstem response” 2019. <em>NeuroImage</em> <strong>200</strong> 1–11</p> <p>[2] Etard O and Reichenbach T. “Neural speech tracking in the theta and in the delta frequency band differentially encode clarity and comprehension of speech in noise” 2019. <em>J. Neurosci.</em> <strong>39</strong> 5750–9</p> <p>[3] Etard O and Reichenbach T. "EEG Dataset for 'Decoding of selective attention to continuous speech from the human auditory brainstem response' and 'Neural Speech Tracking in the Theta and in the Delta Frequency Band Differentially Encode Clarity and Comprehension of Speech in Noise". Doi: 10.5281/zenodo.7086208</p>
Imperial College Ear-EEG Dataset for Auditory Attention Decoding
<p>This repository contains the ear-EEG data and the speech material that were used in our publication "Decoding of Selective Attention to Speech From Ear-EEG Recordings" [1].</p> <p>The data are stored in the archive "icl_earEEG_dataset.zip", which contains:</p> <ul> <li>audio/ - this directory contains the original speech material from each EEG trial.</li> <li>eeg_data.h5: this file contains the unprocessed EEG recordings from each trial, sampled at the original rate of 256 Hz.</li> </ul> <p>The onsets of the EEG and speech material are already aligned - all you need to do is resample the audio (or the features derived therefrom) to the same sample rate as the EEG (256 Hz) in order to obtain time-aligned signals. </p> <p>If you use this dataset for your research, please cite this repository as well as the following article:</p> <p>[1] M. Thornton, D. Mandic, T. Reichenbach, "Decoding of Selective Attention to Speech From Ear-EEG Recordings". <em>submitted</em>. arXiv:2401.05187.</p> <div> <div> <div></div> </div> </div>
ScienceDex guides
Understand access before you commit
These curated guides explain access requirements, typical timelines, costs, and reuse considerations for widely used research datasets.
Allen Brain Atlas
Allen Brain Atlas is an Allen Institute collection of brain map atlases, datasets, APIs, and analysis tools covering mouse, human, and non-human primate brain resources.
Annotated Behaviour and Observability Dataset (ABODe)
ABODe is a University of Edinburgh DataShare dataset for behavior classification in group-housed mice using home-cage video, identities, bounding boxes, ground-plate positions, and annotator labels.
DANDI Archive for NWB datasets
DANDI is a BRAIN Initiative archive for publishing and sharing neurophysiology data, including electrophysiology, optophysiology, and behavioral data packaged as NWB and related standards.
International Brain Laboratory public data
The International Brain Laboratory public data releases expose standardized mouse decision-making experiments, including Neuropixels recordings, widefield calcium imaging, behavior, and session metadata accessed through the ONE API.
OpenNeuro
OpenNeuro is a free, open platform for sharing neuroimaging datasets, with public search, dataset pages, and download paths for web, S3, DataLad, and the OpenNeuro CLI.