Non-native English pronunciation speech data
English as learners really say it: speakers whose first language is Portuguese, Spanish, Japanese, Korean, Arabic or Hindi, reading sentences aloud and answering questions in their own words. English teachers score every recording sound by sound, with word stress and fluency, so pronunciation apps can tell a learner exactly what to fix.
Specification
- Speakers
- Adult learners of English whose first language is Portuguese, Spanish, Japanese, Korean, Arabic or Hindi
- Levels
- From beginner to advanced, with each speaker’s level and first language recorded
- Material
- Read words and sentences, short passages, and spontaneous answers to questions and pictures
- Audio
- 16 kHz, 16-bit WAV, recorded on phones and laptops at home
- Annotation
- The text read, the sounds actually said, each error (substitution, deletion, insertion), word stress, and fluency and accuracy scores per sentence
- Quality check
- Scored by English teachers; a second teacher reviews disagreements and unclear cases
Example of the data
What you see, and the file you get.
| “very” | /ˈvɛri/ → /ˈbɛri/ | v → b · Spanish L1 |
|---|---|---|
| “people” | /ˈpiːpəl/ → /ˈbiːbəl/ | p → b · Arabic L1 |
| “rice” | /raɪs/ → /laɪs/ | r → l · Japanese L1 |
| audio_url | first_language | word | expected | heard | error | score_0_100 |
|---|---|---|---|---|---|---|
| …/audio/rec_0108.wav | es | very | v | b | substitution | 42 |
| …/audio/rec_0108.wav | es | very | ɛ | ɛ | 96 | |
| …/audio/rec_0215.wav | ar | people | p | b | substitution | 38 |
| …/audio/rec_0302.wav | ja | rice | r | l | substitution | 35 |
| …/audio/rec_0302.wav | ja | rice | aɪ | aɪ | 94 | |
| …/audio/rec_0417.wav | pt | think | θ | f | substitution | 40 |
Sample rows. Also as JSON or Excel; audio as WAV or MP3.
Contact us to get the dataset
Tell us what you need. We walk you through the data, its formats and how it is checked on a short call.
What it is used for
Pronunciation coaching apps
Tell a learner which sound went wrong, and how to fix it.
Speaking test scoring
Score read-aloud and spoken answers the way an English teacher would.
Accent-robust speech recognition
Recognise learners and second-language speakers that native-trained models miss.
Mispronunciation detection
Train and benchmark models that find errors sound by sound.
Questions
Who scores the recordings?
English teachers, who mark each sound and word against the text read and give fluency and accuracy scores. Unclear cases go to a second teacher.
Which first languages are included?
Portuguese, Spanish, Japanese, Korean, Arabic and Hindi. Ask us about other first languages you need.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples