Download data and software
1. Software​
All the necessary software tools will be available from Clarity-Cadenza GitHub repository
We recommend installing the software first and then following the instructions in the repository's README for downloading and unpacking the data.
2. Data​
To unpack the data for either of the tasks, we recommend you to follow the instructions in the Clarity-Cadenza Challenge GitHub repository.
All participants will require the core data packages. Participants wishing to extend the training set can use either of the augmentation data packages.
2.1 Task 1 - Headphones​
The data is available for download here.
On the download site you will see five data packages are available,
In case you already have a copy of the MUSDB18-HQ, you won't need to download it from our mirror. However, you still need to request access to the dataset to download the metadata package. This package contains the listeners' audiograms, key information for the HAAQI evaluation.
cadenza_cad1_task1_core_musdb18hq.v1.0.tar.gz[21.1 GB] - core audio data for training, validation and evaluation.cadenza_cad1_task1_core_metadata.v1.0.tar.gz[7 KB] - core metadata for training and validation.
For data augmentation, you can request from their original source:
- MedleyDB version 1: https://zenodo.org/records/1649325 (access upon request)
- MedleyDB version 2: https://zenodo.org/records/1715175 (access upon request)
- BACH10: https://labsites.rochester.edu/air/resource.html (free access)
- FMA_small: https://os.unil.cloud.switch.ch/fma/fma_small.zip (free access)
2.1 Task 2 - Car​
The data is available for download here.
On the download site you will see one data packages available,
cadenza_cad1_task2_core.v1_1.tar.gz[6.5 GB] - core audio and metadata for training and validation.