# Terms and Conditions ## ICASSP 2027 Grand Challenge of EEG-sAAD: Selective Auditory Attention Decoding **Effective date:** September 1, 2026 By registering for and/or participating in the ICASSP 2027 Grand Challenge of EEG-sAAD: Selective Auditory Attention Decoding ("the Challenge"), you ("Participant" or "Team") agree to be bound by the following Terms and Conditions ("Terms"). Please read them carefully. If you do not agree to any of these Terms, you must not register for or participate in the Challenge. --- ### 1. Organizers The Challenge is organized by the EEG-sAAD 2027 Organizing Committee ("Organizers"), comprising members from Nanjing University, KU Leuven, Southern University of Science and Technology (SUSTech), and Anhui University. The Organizers reserve the right to interpret, amend, or supplement these Terms at any time, with notice provided to you. ### 2. Eligibility and Registration Rules 2.1. The Challenge is open to teams from any country or institution worldwide, including academic, industrial, and independent participants. 2.2. Each team must register via the official registration form, providing: **team name, names and affiliations of all members, and a contact email address.** 2.3. Organizers, their immediate family members, and individuals directly involved in the creation, curation, or evaluation of the Challenge (including persons with access to the ground-truth labels of the test sets) are not eligible to participate. 2.4. **A participant may be a member of only one team.** Teams may merge or split only with prior written approval from the Organizers. **The organizers do not guarantee an approval of team merge/split in all cases.** Teams who wish to ask for an approval of merge or split must make sufficient explanation, and the Organizers may require further information prior to the approval. **No merge or split is allowed once Test Phase starts.** 2.5. **The team members should explicitly match the author list of the technical report**, if your team are ranked top-5 in the Test Phase Leaderboard. In any cases of mismatch between team member and the author list, you should provide a reasonable and sufficient formal explanation, acknowledged by all team members, and all authors shown in the technical report. We reserve the right to ask for supplementary materials, or **disqualify your results with an agreement made among all organizers that no sufficient and justified explanation can be made**. In case you wish to add or delete your team members, you should also provide a formal application, acknowledged by all team members. Similar to 2.4, we do not guarantee an approval of addition or deletion of team members. Specifically, in Test Phase, we would require a much more sufficient explanation of your intention of addition or deletion of team members. The intention of these actions is to make sure the competition is as fair as possible. 2.6. **Teams who ranked top-5 in the Test Phase Leaderboard should submit the technical report following ICASSP and IEEE's instructions and requirements.** Any kind of violation to ICASSP and IEEE's submission rules may result in a rejection of your report, and you should take all responsibility. 2.7. There is no registration fee. ### 3. Challenge Description 3.1. The Challenge consists of a **single track**. The task is to decode, from EEG signals and the speech envelopes of two competing talkers, which speaker the listener is attending to at each time point (selective auditory attention decoding, sAAD). 3.2. The Challenge focuses on **cross-dataset generalization**: models are trained on multiple public training datasets and evaluated on held-out blind test sets containing novel subjects and stimuli. ### 4. Data Provision and Usage 4.1. **Training data** (raw+preprocessed EEG, speech envelope, attended speaker label, and attended direction label) will be made available to registered teams via a dedicated cloud drive link, together with baseline code and documentation. 4.2. Training data may be used solely for the purpose of developing, training, and validating models for the Challenge. Any other use, including but not limited to commercial exploitation, redistribution, or public release of the data, is prohibited without prior written permission from the respective data owners. 4.3. **Blind test data** (raw+preprocessed EEG and speech envelope, without labels) will be released via the same channel. Test data may be used **only** for generating Challenge submissions. Specific restrictions are set out in Section 6. 4.4. Test data are distributed with sufficient metadata, for clear expression purpose only. Ground-truth labels are held by the Organizers. Identifiers are provided for file management only and must not be used to improve decoding 4.5. Participants must not attempt to identify subjects, reconstruct identities, or infer any personally identifiable information from any dataset. You are not allowed to improve your decoding results by providing the identity of subject, the trial ID to the decoder. 4.6. Data are provided "as is," without warranty of any kind, express or implied, including but not limited to warranties of accuracy, completeness, or fitness for a particular purpose. ### 5. Ethical and Legal Compliance 5.1. All human EEG data used in the Challenge were collected under appropriate institutional ethical approval. In particular, the NJU-SUSTech test set was recorded with the approval of the SUSTech Ethics Committee (Approval No. 20260006), and all participants provided written informed consent. The KUL test set was recorded with approval of the Medical Ethics Committe UZ Leuven/Research (KU Leuven) with reference S57102, and all participants provided written informed consent. 5.2. Participants must comply with all applicable laws, regulations, and institutional policies in their own jurisdictions, including data protection and privacy legislation (e.g., GDPR, PIPL, HIPAA where applicable). 5.3. Any proposed use of additional external datasets by participants must comply with the licenses and terms of those datasets. Participants are solely responsible for any legal or ethical obligations arising from such use. ### 6. Submission Rules 6.1. You are allowed to submit in both Development Phase and Test Phase. However, only Test Phase results are used for the Leaderboard. 6.1.1. Development Phase submission. Only for the purpose of format checking, we provide the opportunity to you to submit your predictions on the KUL-AV-GC dataset in the last two weeks of the Development Phase. The results are not further used in any other conditions, nor considered as a ranking metric. 6.1.2. Test Phase submission. Each team may submit a maximum of **two (2) submissions** during the Testing Phase; the **latest** submission is used for your ranking. You are **NOT ALLOWED** to submit more than twice with any possible approaches. We reserve the right to disqualify any team that attempts to bypass the submission restrictions with any method, including registration as separate teams. 6.2. Submissions must be made through the designated Challenge platform (CodaBench) before the submission deadline. Late submissions will not be considered. 6.3. **Single forward inference requirement.** Once a model has been trained and validated, the test data, per trial, may only be passed through the model in a single forward inference pass. No data from other test trials or subjects may be used at inference time, only data from the specific test trial, and only following the causality restrictions (see below). No finetuning, unsupervised learning, self-supervised learning, or any other kind of update of model parameters are allowed with test data, except within-trial and if the causality restrictions are satisfied. 6.4. **Causality requirement.** Models must operate with a maximal look ahead of 1 second to produce your submission results. This means that for each provided label for a sample at time t, only samples up to t+1000 ms, from that specific test trial, can be used. An equivalent expression is that, to produce the label at sample t, you are allowed to use samples ranging [0, t+1000ms] at the same test trial and same test subjects. The look-ahead restriction takes all steps in your pipeline into account, including pre- and post-processing steps (disclaimer: the provided pre-processed files were generated using procedures that might have slightly violated the 1s-lookahead rule. We do not take this into account). 6.5. **Code submission for compliance verification.** All teams must submit all code (including training and inference scripts. This code will not be publicly disclosed) alongside their predictions, to enable verification of compliance with these Terms. This does **not** imply a transfer of ownership or IP regarding the submitted code. **The organizers will check compliance** with these rules for at least the top-5 ranked submissions. Teams are highly encouraged to open-source their code after the Challenge concludes. 6.6. Each submission must be reproducible from the submitted code and data, to the reasonable satisfaction of the Organizers. 6.7 The final Leaderboard will include a badge indicating whether a team's submission code has been made publicly available. Since the Organizers will only verify compliance for the top five submissions, the results of other teams are not formally validated. Teams that release their code publicly provide an additional level of transparency, allowing the community to assess compliance and increasing confidence in the credibility of their reported ranking. ### 7. Evaluation and Ranking 7.1 See the `Evaluation` page for detailed descriptions. In summary, the final ranking is determined based on the harmonic mean of decoding accuracy and the switch detection time, on the blind test set. That is to say, only scores in Test Phase Leaderboard will be taken into account. 7.2. The Organizers' evaluation is final and binding. The Organizers reserve the right to adjust rankings or disqualify submissions if compliance issues are identified (Section 9). ### 8. Prizes and Publication 8.1. Up to five (5) top-ranked teams will receive certificates and will be invited to submit a 2-page paper to ICASSP 2027, with potential publication opportunities in the IEEE Open Journal of Signal Processing (OJ-SP). 8.2. Invitation to submit a paper is conditional on the team having complied with all rules, including the submission of model code for compliance verification. 8.3. Publication of any paper derived from the Challenge is subject to the policies and review procedures of IEEE ICASSP and IEEE OJ-SP. The Organizers do not guarantee acceptance of any paper. Participants should appear in the conference site of ICASSP 2027. You are responsible for any kind of violation of the submission rules, author requirements, and any other kind of restrictions of IEEE ICASSP and/or IEEE OJ-SP. ### 9. Disqualification 9.1. A team may be disqualified, at the Organizers' sole discretion, for any of the following: - Violation of any provision of these Terms; - Using test data for training, fine-tuning, or any form of model parameter update; - The usage of future information exceeds the restriction stated in Terms; - Attempting to access ground-truth labels or any held information; - Attempting to improve your score with subject identities, trial ID, or encoding any other held-out information into the input. - Submitting predictions that are not generated by the team's own model or that are derived from the submissions of other teams; - Attempting to manipulate the Leaderboard, evaluation system, or scoring; - Attempting to bypass the submission restrictions with any methods, such as multiple registration; - Failing to submit model code for compliance verification upon request; - Engaging in any conduct that the Organizers reasonably determine to be fraudulent, dishonest, or contrary to the spirit of the Challenge. - Using different model architecture/weight for different test samples. (Note: this rule does not prohibit you from designing a mixture-of-expert decoder. But the proposed decoder, no matter it is a MoE decoder or not, should not, explicitly or implicitly, infer the identity of test samples or utilize such knowledge to make predictions. If such kind of MoE decoder is proposed, it should automatically choose the activated experts solely based on available input data, following the causality restrictions. No manual manipulation of activating or deactivating certain experts is allowed. ) 9.2. In the event of disqualification, the team's rankings, scores, and any prizes or invitations are forfeited. If a top-five team is disqualified, the Organizers may decide at their own discretion whether to move up the next team(s) on the Leaderboard. ### 10. Intellectual Property 10.1. Participants retain ownership of their model code, methods, and algorithms developed for the Challenge. 10.2. By submitting to the Challenge, participants grant the Organizers a non-exclusive, royalty-free, worldwide license to use the submitted predictions, code, and accompanying documentation for the purposes of evaluating the Challenge, verifying compliance, and reporting results. The granted permission will retain if the Organizers will compose a full-length journal of challenge report and submit it to any kind of academic journals, transactions, letter, or other kind of publications. Submitted code will remain confidential, unless the submitting team gives explicit consent to make the code publicly available. 10.3. The datasets are the property of their respective providers. The Organizers are operating this Challenge with granted permission from the respective owners of all datasets used in this Challenge. These Terms do not transfer any ownership rights in the data to participants. 10.4. Nothing in these Terms shall be construed as granting any right to use the name, logo, or trademarks of any participating institution or of IEEE/ICASSP, except as expressly permitted in writing. ### 11. Confidentiality 11.1. Test data, ground-truth labels (until public release), and any evaluation details designated as confidential by the Organizers must be kept confidential by participants. 11.2. Participants must not publicly disclose, discuss, or publish results obtained on the blind test sets before the official results announcement or the public release of test-set labels, whichever occurs later. ### 12. Publicity 12.1. By participating, teams agree that the Organizers may publicly announce team names, affiliations, rankings, and scores in connection with the Challenge. 12.2. Unless expressly authorized, participants may not use the ICASSP or IEEE names or logos in any publicity material without prior written consent. ### 13. Liability 13.1. Participation in the Challenge is voluntary. To the maximum extent permitted by law, the Organizers and their institutions accept no liability for any loss, damage, or injury arising from participation in the Challenge or from the use of the provided data, code, or materials. 13.2. Participants are solely responsible for their own equipment, hardware, software, and any costs incurred in participating. ### 14. Data Protection and Privacy 14.1. The Organizers will process registration information (team name, member names, affiliations, and contact email) solely for the administration and operation of the Challenge, and will not share personal information with third parties except as required for Challenge operation or by law. 14.2. EEG data provided to participants are preprocessed and de-identified; participants must treat all data as confidential and must not attempt re-identification. ### 15. Amendments and Termination 15.1. The Organizers reserve the right to amend these Terms, the Challenge schedule, the evaluation procedure, or to cancel or suspend the Challenge, at any time and for any reason, including technical, legal, or operational reasons. Registered teams will be notified of material changes. 15.2. In the event of cancellation, the Organizers will use reasonable efforts to notify registered teams, but accept no liability for any resulting costs or losses. ### 16. Governing Law and Disputes 16.1. These Terms are governed by and construed in accordance with the laws of the People's Republic of China, without regard to its conflict-of-law principles. 16.2. Any dispute arising out of or in connection with these Terms shall first be attempted to be resolved through friendly consultation between the parties. Failing resolution, the dispute shall be submitted to the competent courts of the place where the lead organizer institution is located. ### 17. Contact Questions regarding these Terms or the Challenge should be directed to the Organizing Committee through the official contact channels provided on the Challenge registration page. --- **By registering for the Challenge, you acknowledge that you have read, understood, and agree to be bound by these Terms and Conditions.**