changelinglab/speechocean-l2eval

1---2license: cc-by-nc-4.03arxiv: 2601.140464dataset_info:5  features:6  - name: audio7    dtype:8      audio:9        sampling_rate: 1600010  - name: speaker_id11    dtype: string12  - name: utt_id13    dtype: string14  - name: text15    dtype: string16  - name: accuracy17    dtype: int3218  - name: completeness19    dtype: float3220  - name: fluency21    dtype: int3222  - name: prosodic23    dtype: int3224  - name: total25    dtype: int3226  splits:27  - name: train28    num_bytes: 26097987429    num_examples: 226030  - name: val31    num_bytes: 3713635832    num_examples: 24033  - name: test34    num_bytes: 28816156735    num_examples: 250036  download_size: 61045312337  dataset_size: 58627779938configs:39- config_name: default40  data_files:41  - split: train42    path: data/train-*43  - split: val44    path: data/val-*45  - split: test46    path: data/test-*47task_categories:48- automatic-speech-recognition49language:50- en51size_categories:52- 1K<n<10K53---54 55# speechocean762: A non-native English corpus for pronunciation scoring task56 57## Dataset Summary58 59**speechocean762** is an open-source non-native English speech corpus designed for **pronunciation assessment** and **L2 spoken proficiency modeling**.60This Hugging Face version provides **sentence-level audio and expert scores**, organized into standard `train` / `validation` / `test` splits.61 62All speakers are Mandarin L1 learners of English, spanning both children and adults. Each utterance is evaluated independently by five expert annotators using standardized pronunciation metrics.63 64This dataset is suitable for:65- pronunciation scoring66- L2 speech assessment67- speech representation learning68- downstream regression or classification tasks69 70## Dataset Structure71 72### Splits73 74The dataset is published with three predefined splits:75- `train` (2260)76- `val` (240)77- `test` (2500)78 79Splits are **speaker-disjoint** and provided as native Hugging Face splits.80 81### Features82 83Each example contains:84 85| Field | Type | Description |86|--------|------|-------------|87| `audio` | `Audio` | Speech waveform (16 kHz) |88| `speaker_id` | `string` | Speaker identifier |89| `utt_id` | `string` | Utterance identifier |90| `text` | `string` | Prompt sentence |91| `accuracy` | `int` | Sentence-level pronunciation accuracy |92| `completeness` | `float` | Percentage of correctly pronounced words |93| `fluency` | `int` | Sentence-level fluency score |94| `prosodic` | `int` | Sentence-level prosody score |95| `total` | `int` | Overall pronunciation score |96 97## Scoring Metrics (Sentence level)98All sentence-level scores follow the original speechocean762 definitions.99For detailed descriptions, see: 100- **arXiv:** https://arxiv.org/abs/2104.01378101- **Github:** https://github.com/jimbozhang/speechocean762102 103## Dataset Creation104This Hugging Face dataset is derived from the original speechocean762 corpus and includes:105- sentence-level audio106- sentence-level expert scores107- standardized HF Audio features108- speaker-disjoint train/val/test splits109 110Word-level and phoneme-level annotations are not included in this version.111 112**Source Dataset**: https://huggingface.co/datasets/mispeech/speechocean762113 114## License115 116The original speechocean762 dataset is released for free use, including commercial and non-commercial purposes, as stated by the original authors.117Users should consult the original repository for full licensing details.118 119## Citation120 121If you use this dataset, please cite the original paper:122 123```bibtex124@inproceedings{zhang2021speechocean762,125  title={speechocean762: An Open-Source Non-native English Speech Corpus For Pronunciation Assessment},126  author={Zhang, Junbo and Zhang, Zhiwen and Wang, Yongqing and Yan, Zhiyong and Song, Qiong and Huang, Yukai and Li, Ke and Povey, Daniel and Wang, Yujun},127  booktitle={Proc. Interspeech 2021},128  year={2021}129}130```131 132## Acknowledgements133 134All credit for data collection and annotation belongs to the original speechocean762 authors.135This Hugging Face release focuses on standardized access and reproducibility for modern speech and representation learning pipelines.136 137You can use this dataset with our benchmarking toolkit at https://github.com/changelinglab/prism138```139@misc{prism2026,140      title={PRiSM: Benchmarking Phone Realization in Speech Models}, 141      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},142      year={2026},143      eprint={2601.14046},144      archivePrefix={arXiv},145      primaryClass={cs.CL},146      url={https://arxiv.org/abs/2601.14046}, 147}148```