IbrahimSalah/The_Arabic_News_speech_Corpus_Dataset

Arabic News Speech Corpus Dataset

This dataset is an Arabic speech corpus that supports the development of syllable-based Arabic speech recognition using Wav2Vec-2 architecture and a 5-gram language model. It consists of Modern Standard Arabic (MSA) syllables extracted from TV news broadcasts, annotated with diacritics.

Dataset Details

Dataset Description

This corpus contains 15 hours of WAV audio recordings transcribed into diacritized Modern Standard Arabic syllables. Initially developed from a 5.7-hour dataset by King Abdulaziz City of Science and Technology (KACST) in 2008, it was later expanded through various grants. The recordings have been processed into syllables using automated software that applies Arabic pronunciation rules. The dataset has a minor error rate (~3%) due to syllabification inaccuracies.

Dataset Sample

'''

Transcription : بَعْ|ضُ|ھُمْ|قَ|لِ|قِمْ|مِنْ|ھَ|ذِ|ھِلْ|خُطْ|وَهْ|وَ|ءَاْ|خَ|رُوْ|نَ|يَرْ|فُ|ضُوْ|نَ|بَ|قَاْ|ءَلْ|قُوْ|وَاْ|تِلْ|ءَمْ|رِيْ|كِيْ|يَهْ|تَحْ|تَ|ءَيْ|يِ|شَرْ|طِنْ|ءَوْ|ظَرْفْ

'''

• Curated by: Moustafa Elshafei
• Language(s): Modern Standard Arabic (MSA)
• License: CC-BY-NC 4.0

Dataset Sources

• Paper: Syllable-Based Arabic Speech Recognition Using Wav2Vec
• DOI: https://doi.org/10.60161/2521-001-001-006

Uses

Direct Use

This dataset is suitable for research and development in automatic speech recognition (ASR) systems for Arabic, particularly those based on syllables and diacritics. It can be used to train deep learning models, such as Wav2Vec-2, for syllable recognition and speech-to-text conversion in MSA.

Out-of-Scope Use

Commercial use is prohibited under the CC-BY-NC 4.0 license.

Dataset Structure

The dataset is split into training and testing (1.5 hours) subsets. Each audio file is paired with its corresponding syllable transcription.

Dataset Creation

Curation Rationale

This dataset was created to address the challenges of Arabic speech recognition, particularly the complexities of syllable-based pronunciation and diacritization.

Source Data

The data is collected from Arabic news broadcasts and processed using automated tools that convert MSA text into syllables. The source dataset initially contained 5.7 hours of recordings but has since been expanded.

Data Collection and Processing

• The recordings were converted into syllables using an automated tool based on Arabic pronunciation rules.
• Diacritized syllables provide a more accurate representation of speech variations.

Who are the source data producers?

The source data comes from TV news broadcasts, curated and processed by the researchers at Zewail City of Science and Technology.

Annotations

The dataset contains syllable-level annotations, which include diacritics for precise pronunciation mapping.

Annotation process

Annotations were generated automatically using pronunciation rules for MSA, though a small error rate is acknowledged.

Who are the annotators?

Annotations were generated through automated tools rather than human annotators.

Personal and Sensitive Information

This dataset does not contain any personal or sensitive information.

Bias, Risks, and Limitations

The automated nature of syllable processing introduces a minor error rate (estimated at 3%). The dataset is primarily intended for research and academic use and may not perform optimally in real-world commercial applications without further refinement.

Recommendations

For academic research, users should be aware of potential syllabification inaccuracies, especially in edge cases where foreign words are forced to follow Arabic syllable patterns.

Citation

BibTeX:

@article{2024SyllableBasedAS,
  title={Syllable-Based Arabic Speech Recognition Using Wav2Vec},
  author={إبراهيم عبدالعال and مصطفى الشافعي and محمد عبدالواحد},
  journal={مجلة اللغات الحاسوبية والمعالجة الآلية للغة العربية},
  year={2024},
  url={https://api.semanticscholar.org/CorpusID:269151543}
}