changelinglab/l2arcticperceived-pr

L2-ARCTIC (Perceived)

[!NOTE] This dataset is licensed under CC BY-NC 4.0. If you use this dataset, please remember to cite: ``` @inproceedings{zhao2018l2arctic, % curated the dataset

title={L2-ARCTIC: A Non-native English Speech Corpus}, author={Zhao, Guanlong and Sonsaat, Sinem and Silpachai, Alif and Lucic, Ivana and Chukharev-Hudilainen, Evgeny and Levis, John and Gutierrez-Osuna, Ricardo}, booktitle={Proc. Interspeech 2018}, pages={2783--2787}, year={2018} }

@inproceedings{zhu2025zipa, % processed this subset

title = "{ZIPA}: A family of efficient models for multilingual phone recognition", author = "Zhu, Jian and Samir, Farhan and Chodroff, Eleanor and Mortensen, David R.", booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2025", address = "Vienna, Austria", publisher = "Association for Computational Linguistics", doi = "10.18653/v1/2025.acl-long.961", }

```

L2-ARCTIC is a speech corpus of non-native English from 24 speakers, providing detailed transcription, alignment, and error information on different level.

This is a subset that only includes manual annotations, i.e., human transcriptions of the actual pronunciation. The speech is resampled from 44.1kHz to 16kHz, while the phones are extracted from TextGrid and mapped from ARPAbet to IPA.

This dataset is included as pr-arc in our phone recognition benchmarking toolkit, 💎PRiSM.

@misc{prism2026,
      title={PRiSM: Benchmarking Phone Realization in Speech Models}, 
      author={Shikhar Bharadwaj and Chin-Jou Li and Yoonjae Kim and Kwanghee Choi and Eunjung Yeo and Ryan Soh-Eun Shim and Hanyu Zhou and Brendon Boldt and Karen Rosero Jacome and Kalvin Chang and Darsh Agrawal and Keer Xu and Chao-Han Huck Yang and Jian Zhu and Shinji Watanabe and David R. Mortensen},
      year={2026},
      eprint={2601.14046},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2601.14046}, 
}