IWAN/L2-KSU-Dataset

L2-KSU – Arabic Mispronunciation Detection and Diagnosis (MDD) Dataset

📄 Paper: Improving mispronunciation detection and diagnosis for non-native learners of the Arabic language — Discover Computing 28(1):1, 2025

L2-KSU is an Arabic speech corpus for mispronunciation detection and diagnosis (MDD), collected at King Saud University. It contains read-speech recordings from both native and non-native (L2) speakers of Arabic (male and female), with phone-level annotations for diagnosing pronunciation errors.

Contents

The audio corpus is distributed as a single archive, L2-KSU-Dataset.zip, with this structure:

L2-KSU-Dataset/
├── native_speakers/   {males,females}/speakerN/sentenceN/{wav,transcript,annotation}/
└── non-native_speakers/{males,females}/speakerN/sentenceN/{wav,transcript,annotation}/

File type	Count	Description
.wav	4,094	Audio recordings (one per utterance)
.TextGrid	4,098	Praat phone-level annotations
.txt	4,174	Per-utterance transcripts

annotations.csv

A flat table of 818 annotated utterances used for the MDD experiments.

Column	Description
text	Reference sentence, word-level (ground truth)
canon	Canonical reference, character/phone-level (ground truth)
ans	Actual pronounced sequence, character/phone-level (captures real pronunciation errors)
path	Original path of the audio file (rewrite to your local copy of L2-KSU-Dataset.zip)

Note: The original package shipped train_set.csv and test_set.csv that were byte-for-byte identical, so a single annotations.csv is provided here. No official train/test split is included — define your own split as needed.

Citation

If you use L2-KSU, please cite:

@article{alrashoudi2025improving,
  title     = {Improving mispronunciation detection and diagnosis for
               non-native learners of the Arabic language},
  author    = {Alrashoudi, Norah and Al-Khalifa, Hend and Alotaibi, Yousef},
  journal   = {Discover Computing},
  volume    = {28},
  number    = {1},
  pages     = {1},
  year      = {2025},
  publisher = {Springer},
  doi       = {10.1007/s10791-024-09489-8},
  url       = {https://doi.org/10.1007/s10791-024-09489-8}
}