speechocean762: non-native English pronunciation scoring corpus
- Provider
- Junbo Zhang et al. (Xiaomi, SpeechOcean)
- License
- CC-BY-4.0 (source)
- Size
- 4 files
- Format
- Parquet (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/mispeech/speechocean762)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Speech & Audio
- Source
- https://arxiv.org/abs/2104.01378
About speechocean762: non-native English pronunciation scoring corpus
speechocean762 is an open corpus for automatic pronunciation assessment in computer-assisted language learning. It consists of 5,000 English sentences read by non-native speakers whose mother tongue is Mandarin; half of the speakers are children and half adults, with age and gender provided. Five human experts scored every utterance independently under the same rubric, giving manual annotations at sentence level (accuracy, completeness, fluency, prosody, total), word level (accuracy, stress, total) and phoneme level (per-phone accuracy and marked mispronunciations). It is the standard benchmark for goodness-of-pronunciation (GOP) and pronunciation scoring models and for mispronunciation detection and diagnosis. This repository holds the corpus as two Parquet files under data/ (train and test, 2,500 utterances each per the card example), with the audio and nested word and phone annotations (measured 2026-11-14; the Hub reports about 2 GB). Load with datasets.load_dataset('mispeech/speechocean762', split='test') or pandas. Released on OpenSLR (SLR101) under CC BY 4.0 for commercial and non-commercial use; cite Zhang et al., Interspeech 2021.
How to load speechocean762: non-native English pronunciation scoring corpus in Python
from datasets import load_dataset
ds = load_dataset("mispeech/speechocean762")
print(ds)Data structure
File types in sampled files: Parquet
Files
Example file: data/test-00000-of-00001.parquet
Analyze this dataset without downloading it
On Scigantic, speechocean762: non-native English pronunciation scoring corpus opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Speech & Audio datasets
- Upper Waiakea Forest Reserve Passive Acoustic Recordings of Hawaiian Forest Birds
- 2 M Belebele
- WAXAL: multilingual African language speech corpus (ASR and TTS)
- YODAS v3: 1.1-million-hour multilingual web speech corpus
- Seamless Interaction
- 2 M Flores ASL
- Granary
- Crowdsourced UK and Ireland English Dialect Speech (OpenSLR 83)