This dataset takes the open-asr-leaderboard/datasets-test-only data and sorts each split by audio length.
The format is also changed, from custom loading script (un-safe remote code) to parquet (safe).
Broadly speaking, this dataset was generated with the following code-snippet:
from datasets import load_dataset, get_dataset_config_names
DATASET = "open-asr-leaderboard/datasets-test-only" # dataset to load from
HUB_DATASET_ID =… See the full description on the dataset page:
https://huggingface.co/datasets/hf-audio/open-asr-leaderboard.
