Skip to main content

Submission

Registration​

All teams are required to register. Registered teams will be assigned a unique team ID to identify their submission. Please mind that you will receive your TEAM_ID after the submission windows open. This will be sent to the same email account you use during registration.

Evaluation data?​

The evaluation data will have the same format as the validation data (see details in data specification). Pairs of processed and unprocessed audio signals will be provided together with their metadata using the same fields as the validation dataset. It won't contain the listener's correctness score or their responses. We will score your submission for you and return your score (we aim to do this within 48 hours of submission).

release of the Evaluation Data

See Key Dates to check when evaluation data will be released.

What do I need to submit?​

All teams must submit

  • CSV file with the evaluation predictions (evaluation set) with scores ranging between 0 and 100.
  • CSV file with the validation prediction (validation set) with scores ranging between 0 and 100.
  • A two-page technical report
warning

Please note: we can only score your submission if it includes predictions for both the evaluation and validation sets.

The predicted intelligibility scores​

The predicted scores need to be submitted in a CSV file named ICASSP2027_<DATA SET>_<TEAMS_ID>.csv where:

  • <TEAM_ID> is your individual TEAM_ID provided after registration,
  • <DATA SET> is either valid or eval for validation and evaluation sets, respectably.

For example, for team T001, the files with the predicted scores should be called:

  • ICASSP2027_valid_T001.csv for the validation set predictions.
  • ICASSP2027_eval_T001.csv for the evaluation set predictions.

The CSV files should have two columns with headers,

signal_id, intelligibility_score

where the signal_id is the unique signal identifier used for each FLAC file name (the metadata's signal field) and the intelligibility_score is the predicted intelligibility given in terms of the percentage words recognised correctly for the signal (i.e. between 0 and 100).

The technical report​

  • A two-page technical report must be submitted alongside your predicted scores.
  • Your report should include an abstract and introduction and sections on experimental setup/methodology including system information and model/network architecture, evaluation/results, discussion, conclusion and references. Please provide an estimation of the computational resources needed. You must describe any pre-existing tools, software and models used.
  • We suggest you use the official ICASSP 2027 Paper Kit.
  • The top five systems will be invited to submit a paper to the ICASSP 2027 special session.

Note, you will not have your final evaluation set scores when you submit your report. We will score your submission for you and return your score (we aim to do this within 48 hours of submission). This extra information can then be included in a revised version of your report, which will be published on our website in time for ICASSP 2027 on May.

Where should I submit my entry?​

The CSV files and the technical report should be submitted using the following form:
Submission Form

Multiple submissions​

If your team wishes to make more than one submission, then please prepare separate CSV files adding 'a', 'b', etc to your <TEAM_ID>, e.g. if your <TEAM_ID> is T001 use the names ICASSP2027_valid_T001a.csv and ICASSP2027_valid_T001b.csv for the validation set. Same structure for the evaluation set. We will only accept multiple entries where we judge the systems to be significantly different, i.e. separate entries cannot be the same system with different hyperparameter tunings.

If you have submitted multiple entries, you may describe them both in the same technical report or use separate technical reports if the systems are unrelated, and you need more space.

Code​

We encourage you to make your code open source.

How will intellectual property be handled?​

See Intellectual Property section for details.