Hugging Face's logo

Datasets:
refine-ai
/
subscene

Tasks:
Text Generation
Translation
Text Classification
Languages:
Arabic
Armenian
Azerbaijani
Size:
1B<n<10B
Dataset card Files Community
9
Dataset Viewer

The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Subscene

Subscene: The largest collection of subtitles

Dataset Summary

Subscene is a vast collection of multilingual subtitles, encompassing 65 different languages and consisting of more than 30 billion tokens with a total size of 410.70 GB. This dataset includes subtitles for movies, series, and animations gathered from the Subscene dump. It provides a rich resource for studying language variations and building multilingual NLP models. We have carefully applied a fastText classifier to remove any non-language content from incorrect subsets. Additionally, we performed basic cleaning and filtration. However, there is still room for further cleaning and refinement.

Dataset Structure

Data Instances 

{"subtitle_name": "choir-girl",
"file_name": "Choir.Girl.2019.HDRip.XviD.AC3-EVO.json",
"transcript": [{"id": 1, "start_time": "00:00:09,741", "end_time": "00:00:13,702", "text": "Subtitles by explosiveskull"}, {"id": 2, "start_time": "00:01:58,569", "end_time": "00:02:00,536", "text": "Get Julius on the phone."}, {"id": 3, "start_time": "00:02:00,538", "end_time": "00:02:03,338", "text": "The sponsor's got an issue with sharing the page."}, {"id": 4, "start_time": "00:02:08,412", "end_time": "00:02:09,411", "text": "Eugene."}, {"id": 5, "start_time": "00:02:10,748", "end_time": "00:02:14,950", "text": "Yeah, uh, I'm supposed to meet Julius."}, {"id": 6, "start_time": "00:02:14,952", "end_time": "00:02:19,888", "text": "I'm Anne Kellan, the editor. You won't be meeting him."}, {"id": 7, "start_time": "00:02:19,890", "end_time": "00:02:22,391", "text": "He said on the phone that, um..."}, {"id": 8, "start_time": "00:02:22,393", "end_time": "00:02:24,293", "text": "We won't be publishing these."}, {"id": 9, "start_time": "00:02:24,295", "end_time": "00:02:31,266", "text": "You have a child overdosing, an elderly woman being bashed"}, {"id": 10, "start_time": "00:02:31,268", "end_time": "00:02:33,302", "text": "and this,"}, {"id": 11, "start_time": "00:02:33,304", "end_time": "00:02:36,205", "text": "a girl either being sexually assaulted or prostituting herself."}, {"id": 12, "start_time": "00:02:36,907", "end_time": "00:02:39,374", "text": "I spoke to Julius and..."}, ...]
}

{"subtitle_name": "the-fabulous-first-season",
 "file_name": "The.Fabulous.S01E06.DUBBED.WEBRip.x264-ION10_Arabic.json",
 "transcript": [{"id": "1", "start_time": "00:00:06,006", "end_time": "00:00:09,926", "text": "‫\"مسلسلات NETFLIX\"‬"}, {"id": "2", "start_time": "00:00:18,309", "end_time": "00:00:19,849", "text": "‫- المجال مفتوح أمامك.‬ ‫- اركض!‬"}, {"id": "3", "start_time": "00:00:19,936", "end_time": "00:00:22,186", "text": "‫تلك فرصتك! هيا يا \"وو مين\"!‬"}, {"id": "4", "start_time": "00:00:23,189", "end_time": "00:00:24,229", "text": "‫- ويلاه. ‬ ‫- لا بأس.‬"}, {"id": "5", "start_time": "00:00:24,315", "end_time": "00:00:26,025", "text": "‫رائع، أحسنت صنعًا!‬"}, {"id": "6", "start_time": "00:00:26,109", "end_time": "00:00:27,189", "text": "‫أحسنت!‬"}, {"id": "7", "start_time": "00:00:27,277", "end_time": "00:00:28,397", "text": "‫ألن تخبرني؟‬"}, {"id": "8", "start_time": "00:00:29,320", "end_time": "00:00:30,490", "text": "‫الفتاة التي في الصورة.‬"}, {"id": "9", "start_time": "00:00:31,072", "end_time": "00:00:32,622", "text": "‫كم هو مثابر. العب الكرة فحسب.‬"}, ....]
}

Data Fields

  • subtitle_name (str): The name of the folder containing the subtitle file. This likely corresponds to the movie, series, or anime title.
  • file_name (str): The specific filename of the subtitle file itself.
  • transcript (dict): This dictionary holds the complete transcript information for the subtitle file. It contains nested structures to represent individual subtitle segments.
  • id (int): (Within transcript dictionary) A unique identifier for a specific subtitle segment within the file.
  • start_time (str): (Within transcript dictionary) The starting time of the subtitle segment, likely formatted according to a specific timecode standard.
  • end_time (str): (Within transcript dictionary) The ending time of the subtitle segment, following the same timecode format as start_time.
  • text (str): (Within transcript dictionary) The actual text content displayed for that specific subtitle segment.

Using The HuggingFace datasets 

from datasets import load_dataset

# Load the Arabic subset only
ds = load_dataset("refine-ai/Subscene", "arabic", trust_remote_code=True)

Dataset Statstics

Language Raw Size Num Char Num Byte Num Words Num Token
Arabic 59.7 GB 12B 24B 3B 5B
Armenian 933.7 KB 175K 372K 36K 177K
Azerbaijani 5.7 MB 1M 2M 225K 701K
Basque 36.4 MB 9M 10M 1M 3M
Belarusian 3.2 MB 598K 1M 119K 433K
Bengali 2.6 GB 429M 1B 88M 152M
traditional_chinese 1.1 GB 202M 323M 38M 110M
Bosnian 51.0 MB 12M 15M 3M 6M
Brazilian-Portuguese 9.1 GB 2B 3B 463M 727M
Bulgarian 324.7 MB 60M 126M 13M 38M
Bulgarian-English 4.2 MB 851K 1M 179K 515K
Burmese 724 MB 127M 378M 11M 237M
Cambodian-Khmer 165.5 MB 28M 80M 2M 32M
Catalan 70.8 MB 17M 20M 4M 6M
simplified_chinese 3.9 GB 197M 505M 29M 129M
Croatian 1.1 GB 260M 315M 54M 129M
Czech 2.4 GB 519M 671M 108M 288M
Danish 12.1 GB 3B 4B 716M 1B
Dutch 5.5 GB 1B 2B 319M 610M
Dutch-English 28.3 MB 7M 9M 2M 3M
English 127.9 GB 30B 36B 7B 10B
English-German 160.5 MB 39M 47M 8M 14M
Esperanto 4.6 MB 1M 1M 224K 431K
Estonian 345.9 MB 83M 98M 15M 39M
Farsi_persian 40.3 GB 7B 15B 2B 5B
Finnish 4.3 GB 1B 1B 179M 522M
French 18 GB 5B 6B 1B 1B
Georgian 2.3 MB 432K 1M 79K 378K
German 3 GB 788M 931M 157M 307M
Greek 4.5 GB 919M 2B 181M 737M
Greenlandic 10.2 MB 3M 3M 298K 1M
Hebrew 4.9 GB 839M 2B 194M 811M
Hindi 575 MB 92M 285M 24M 36M
Hungarian 1.2 GB 287M 353M 52M 141M
Hungarian_English 600kb 126K 152K 25K 53K
Icelandic 476.1 MB 110M 144M 23M 58M
Indonesian 9.7 GB 10B 12B 2B 3B
Italian 7 GB 2B 3B 501M 950M
Japanese 2.0 GB 198M 596M 29M 190M
Kannada 21.8 MB 4M 11M 564K 1M
Korean 5.5 GB 553M 2B 188M 821M
Kurdish 72.4 MB 14M 29M 3M 11M
Latvian 157.2 MB 35M 43M 6M 18M
Lithuanian 207.3 MB 47M 57M 8M 24M
Macedonian 97 MB 20M 41M 4M 12M
Malay 4.9 GB 1B 1B 221M 371M
Malayalam 446.9 MB 86M 246M 11M 27M
Manipuri 1000kb 93K 280K 14K 60K
Mongolian 9.9 MB 2M 4M 390K 2M
Nepali 37.8 MB 6M 19M 1M 2M
Norwegian 6.8 GB 2B 2B 392M 765M
Pashto 9.5 MB 2M 4M 417K 870K
Polish 2 GB 467M 570M 85M 243M
Portuguese 3.7 GB 899M 1B 191M 280M
Punjabi 55.3 MB 87K 263K 22K 33K
Romanian 3.5 GB 835M 1B 183M 402M
Russian 1.9 GB 359M 740M 71M 220M
Serbian 929.5 MB 225M 285M 48M 113M
Sinhala 1.1 GB 185M 544M 37M 345M
Slovak 649.2 MB 140M 179M 29M 75M
Slovenian 654.8 MB 151M 183M 31M 74M
Somali 1.1 MB 300K 353K 62K 134K
Spanish 10.9 GB 3B 3B 548M 831M
Sundanese 3.1 Mb 762K 881K 135K 302K
Swahili 15.9 MB 4M 4M 661K 1M
Swedish 6.9 GB 2B 2B 392M 795M
Tagalog 188 MB 45M 52M 9M 18M
Tamil 1 GB 50M 149M 7M 16M
Telugu 223.5 MB 39M 115M 6M 14M
Thai 6.9 GB 1B 3B 77M 1B
Turkish 6.7 GB 2B 2B 268M 789M
Ukrainian 891.1 MB 168M 346M 34M 116M
Urdu 5 GB 53M 116M 15M 21M
Vietnamese 16 GB 3B 5B 952M 1B
Yoruba 820.6 KB 154K 230K 46K 80K

Citation Information

If you use this corpus, please cite the paper:

@misc{refine-ai,
author = {Refine AI},
title = {Subscene: A Large-Scale Multilingual Subtitle Dataset},
year = {2024},
}

Acknowledgements

We would like to extend our thanks to everyone who has worked on and contributed to the Subscene website. A special thanks goes to eionlol for uploading the entire Subscene dump, organized by language. You can access the archive through the following link: https://archive.org/details/subscene-final-dump

Downloads last month
2,656