ufal/IWSLT2024_Levantine_Arabic_data

IWSLT2024 - Dialectal and Low-resource speech translation track: North Levantine Arabic # IWSLT2024 - Dialectal and Low-resource speech translation track: North Levantine Arabic Repository for sharing the audio and textual data in the North Leventine Arabic language (ISO-3 code: _apc_), one of the languages for the low-resource speech translation track at IWSLT 2024. ## Training data Please see the IWSLT website. ## Validation data ~~[**17.02.2024**] We fixed the incorrect timestamps for two dialogues: __Tar_13052022_Food__ and __Tar_13052022_Czechia__. Both `segments` and `valid.yaml` files were effected, `valid.{apc,eng}` have not changed.~~ ~~We follow the same format as the previous shared task editions, see `data/valid/txt`. The audio files can be downloaded from here. When prompted for password, use __lowercased name of this repository__, e.g., `github.com -> foo_bar_1`.~~ The data corresponding to the IWSLT 2024 validation split is available here. In the `segments` file, we follow the `offset ||| duration` formulation, formating the timestamps in the `"%S.%03d"` format, i.e., seconds and miliseconds, see the `valid.yaml`. ## 2024 Test data ~~[01.04.2024] We have released the segments file corresponding to 2024 test data, see `data/test2024/txt`. The audio files can be now downloaded from the same source as validation data, using the same password.~~ The data corresponding to the IWSLT 2024 test split is available here. ## 2025 Test data ~~[07.04.2025] We have released the segments file corresponding to 2025 test data, see `data/test2025/txt`. The audio files can be now downloaded here. When prompted for password, use __lowercased name of this repository__, e.g., `github.com -> foo_bar_1`.~~ ~~[18.04.2025] If you are unable to download the audio data from the link above, please use this one. The password is the same, i.e __lowercased name of this repository__, e.g., `github.com -> foo_bar_1`.~~ The data corresponding to the IWSLT 2025 test split is available here. ## License The data is p …