linagora/linto-dataset-audio-ar-tn

LinTO DataSet Audio for Arabic Tunisian
A collection of Tunisian dialect audio and its annotations for STT task

This is the first packaged version of the datasets used to train the Linto Tunisian dialect with code-switching STT (linagora/linto-asr-ar-tn).

• Dataset Summary
• Dataset composition
  • Sources
  • Data Table
  • Data sources
  • Content Types
  • Languages and Dialects
• Example use (python)
• License
• Citations

Dataset Summary

The LinTO DataSet Audio for Arabic Tunisian is a diverse collection of audio content including music, documentaries, podcasts, and other types of recordings, along with their corresponding transcriptions. The dataset is primarily focused on supporting speech recognition tasks for the Tunisian dialect, with some instances of code-switching between Tunisian Arabic, French, and English. It is organized into multiple configurations and splits for different experimental setups, making it valuable for Automatic Speech Recognition (ASR) research and development.

Dataset Composition

The LinTO DataSet Audio for Arabic Tunisian comprises a diverse range of audio content collected from multiple sources. Below is a breakdown of the dataset’s composition:

Sources

• Hugging Face Datasets: Various datasets obtained from the Hugging Face platform.
• YouTube: Audio collected from a range of YouTube channels and videos, including both shorts and long-form content, covering music, documentaries, and podcasts.
• Websites: Audio gathered from various online sources, including educational sites and story-sharing platforms.

Data Table

subset	audio duration	labeled audio duration	# audios	# segments	# words	# characters
AmenyKH	4h 5m 28s + 3m 42s	4h 5m 28s + 3m 42s	3094 + 44	3094 + 44	31713 + 483	158851 + 2462
ApprendreLeTunisien	37m 30s + 3m 4s	37m 31s + 3m 4s	878 + 116	878 + 116	1148 + 156	5220 + 711
MASC	2h 52m 43s	1h 37m 11s	47	1728	11750	59013
OneStory	1h 32m 47s + 8m 3s	1h 31m 15s + 7m 43s	36 + 3	494 + 43	12336 + 1028	56878 + 5059
TunSwitchCS	10h 0m 56s + 27m 17s	10h 0m 56s + 27m 17s	5377 + 300	5377 + 300	74947 + 4253	391038 + 22304
TunSwitchTO	3h 19m 6s + 28m 38s	3h 19m 6s + 28m 38s	2195 + 344	2195 + 344	18222 + 2736	94020 + 14102
Tunisian_dataset_STT-TTS15s_filtred1.0	3h 49m 42s	3h 49m 42s	1029	1029	33428	172927
Wav2Vec-tunisian-Darja	3h 26m 33s	3h 26m 33s	7898	7898	20352	104176
Youtube_AbdelAzizErwi	24h 34m 10s	21h 54m 32s	25	21940	131544	623434
Youtube_BayariBilionaire	58m 50s	55m 5s	6	1080	7813	39831
Youtube_DiwanFM	5h 27m 6s	4h 2m 43s	36	4670	30310	152352
Youtube_HamzaBaloumiElMohakek	16h 41m 52s	14h 5m 49s	21	13734	89334	446736
Youtube_HkeyetTounsiaMensia	1h 44m 47s	1h 24m 46s	5	1518	10528	51570
Youtube_LobnaMajjedi	57m 22s	53m 13s	2	886	6134	30216
Youtube_MohamedKhammessi	1h 43m 53s	1h 34m 3s	2	1825	13216	64141
Youtube_Qlm	2h 31m 33s	1h 51m 34s	53	2541	15728	83682
Youtube_TNScrapped_V1	4h 8m 02s + 18m 42s	2h 33m 30s + 9m 53s	52 + 5	2538 + 179	18777 + 1448	92531 + 7375
Youtube_TN_Shorts	3h 46m 38s	3h 23m 38s	135	2022	28129	143966
Youtube_TV	36m 37s	31m 34s	4	668	4768	24006
TOTAL	92h 55m 48s / 1h 29m 26s	81h 38m 09s / 1h 20m 17s	20895 / 812	76115 / 1026	560177 / 10104	2794588 / 52013

NB: The + in each information column indicates the combined train + test data. For any datasets other than YouTube, which include their links, please review the provided links for additional details.

Data Proccessing:

• Audio Alignment: Matching audio segments with corresponding text to ensure accurate transcription and contextual alignment.
• Transcription Correction: Reviewing and correcting transcriptions to address errors and discrepancies in the initial text.
• Standardization: Converting words and phrases into their standardized forms to maintain consistency across the dataset.
• Padding: Adding padding to shorter audio segments to address issues with Kaldi and ensure uniformity in input lengths.
• Silence Removal: Eliminating segments of audio that contain only silence to improve dataset efficiency and relevance.
• Annotation: Labeling audio segments that require transcriptions and other metadata. Ensuring that non-annotated audio is reviewed and annotated if necessary.

Content Types

• Music: Includes recordings of different music genres.
• FootBall: Includes recordings of football news and reviews.
• Documentaries: Audio from documentaries about history and nature.
• Podcasts: Conversations and discussions from various podcast episodes.
• Authors: Audio recordings of authors reading or discussing different stories: horror, children's literature, life lessons, and others.
• Lessons: Learning resources for the Tunisian dialect.
• Others: Mixed recordings with various subjects.

Languages and Dialects

• Tunisian Arabic: The primary focus of the dataset, including Tunisian Arabic and some Modern Standard Arabic (MSA).
• French: Some instances of French code-switching.
• English: Some instances of English code-switching.

Characteristics

• Audio Duration: The dataset contains approximately 93 hours of audio recordings.
• Segments Duration: This dataset contains segments, each with a duration of less than 30 seconds.
• Labeled Data: Includes annotations and transcriptions for a significant portion of the audio content.

Data Distribution

• Training Set: Comprises a diverse range of audio recordings, each representing different contexts, aimed at enhancing the model's performance across various scenarios.
• Testing Set: onsists of a varied set of audio recordings, also covering different contexts, dedicated to assessing the model’s performance and generalization.

This composition ensures a comprehensive representation of various audio types and linguistic features, making the dataset valuable for a range of ASR research and development tasks.

Example use (python)

• Load the dataset in python:

from datasets import load_dataset

# dataset will be loaded as a DatasetDict of train and test
dataset = load_dataset("linagora/linto-dataset-audio-ar-tn")

Check the containt of dataset:

example = dataset['train'][0] 
audio_array = example['audio']["array"]
segments = example['segments']
transcription = example['transcript']

print(f"Audio array: {audio_array}")
print(f"Segments: {segments}")
print(f"Transcription: {transcription}")

Example

Audio array: [0. 0. 0. ... 0. 0. 0.]
Transcription: أسبقية قبل أنا ما وصلت خممت فيه كيما باش نحكيو من بعد إلا ما أنا كإنطريبرنور كباعث مشروع صارولي برشا مشاكل فالجستين و صارولي مشاكل مع لعباد لي كانت موفرتلي اللوجسيل ولا اللوجسيل أوف لنيه ولا لوجسيل بيراتي
segments: [{'end': 14.113, 'start': 0.0, 'transcript': 'أسبقية قبل أنا ما وصلت خممت فيه كيما باش نحكيو من بعد إلا ما أنا كإنطريبرنور كباعث مشروع صارولي برشا مشاكل فالجستين و صارولي مشاكل مع لعباد لي كانت موفرتلي اللوجسيل ولا اللوجسيل أوف لنيه ولا لوجسيل بيراتي', 'transcript_raw': 'أسبقية قبل أنا ما وصلت خممت فيه كيما باش نحكيو من بعد إلا ما أنا كإنطريبرنور كباعث مشروع صارولي برشا مشاكل فالجستين و صارولي مشاكل مع لعباد لي كانت موفرتلي اللوجسيل ولا اللوجسيل أوف لنيه ولا لوجسيل بيراتي'}]

License

Given that some of the corpora used for training and evaluation are available only under CC-BY-4.0 licenses, we have chosen to license the entire dataset under CC-BY-4.0.

Citations

When using the LinTO DataSet Audio for Arabic Tunisian corpus, please cite this page:

@misc{linagora2024Linto-tn,
  author = {Hedi Naouara and Jérôme Louradour and Jean-Pierre Lorré},
  title = {LinTO Audio and Textual Datasets to Train and Evaluate Automatic Speech Recognition in Tunisian Arabic Dialect},
  year = {2024},
  month = {October},
  note = {Good Data Workshop, AAAI 2025},
  howpublished = {\url{https://huggingface.co/linagora/linto-asr-ar-tn-0.1}},
}

@misc{abdallah2023leveraging,
      title={Leveraging Data Collection and Unsupervised Learning for Code-switched Tunisian Arabic Automatic Speech Recognition}, 
      author={Ahmed Amine Ben Abdallah and Ata Kabboudi and Amir Kanoun and Salah Zaiem},
      year={2023},
      eprint={2309.11327},
      archivePrefix={arXiv},
      primaryClass={eess.AS}
}

@data{e1qb-jv46-21,
doi = {10.21227/e1qb-jv46},
url = {https://dx.doi.org/10.21227/e1qb-jv46},
author = {Al-Fetyani, Mohammad and Al-Barham, Muhammad and Abandah, Gheith and Alsharkawi, Adham and Dawas, Maha},
publisher = {IEEE Dataport},
title = {MASC: Massive Arabic Speech Corpus},
year = {2021} }