All the necessary software tools are available as a single GitHub repository.
A new version of the repository code, v0.2.0, has been released for use with the challenge. It contains a
recipe for running the baseline and standard evaluation (
The data is available for download here.
On the download site you will see four data packages are available,
clarity_CEC2_core.v1_1.tgz[28 GB] - metadata and dev set
clarity_CEC2_train.v1_1.tgz[69 GB] - scenes for training systems
clarity_CEC2_hoairs.v1_0.tgz[144 GB] - impulse responses for generating extended training data
clarity_CEC2_icassp2023_eval.v1_0.tgz[6.5 GB] - the
eval2evaluation sets (added on 2nd Feb 2023).
All participants will require the core data package. Participants using machine learning approaches will additionally require the train data package. Participants wishing to extend the training set by using our provided scene rendering tools will also require the high order ambisonic impulse responses (i.e., the hoairs package).
To unpack the data we recommend you follow the instructions in the Clarity Challenge GitHub repository.