Dataset · mispeech
English Speech Audio Corpus for Pronunciation Assessment
mispeech/speechocean762
Non-native English speech recordings with pronunciation quality annotations at the phoneme, word, and sentence levels.
What's inside
Contains 1k to 10k audio recordings of non-native English speech along with text transcripts. The standout feature is pronunciation quality scores provided at three distinct levels: individual phonemes, whole words, and full sentences. Created specifically for Computer-Assisted Language Learning (CALL) systems.
What you can do with it
- Training and testing automated pronunciation evaluation models in language apps
- Fine-tuning ASR systems for improved recognition of accented speech
- Building algorithms for detailed feedback on phonetic errors
- Tasks
- Automatic speech recognition
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 3.2K
- Likes
- 31
Models trained on it
- tvosch/Qwen3-ASR-0.6B-Jev-DutchAudio classification 315
- vb223/whisper-base-en-phoneme-ctc 261
- anonym-coling2027/E2E-GOP-SpeechOcean762 184
- OpenPronounce/wav2vec2-base-phones-enAutomatic speech recognition 86
- saugataroy/ielts-speaking-scorer 17
- gyrroa/whisper-custom-smallAutomatic speech recognition 16
- Pooya-Fallah/whisper-tiny-finetuneAutomatic speech recognition 10
- stringbot/whisper-small-hiAutomatic speech recognition 3
How to load it
from datasets import load_dataset
ds = load_dataset("mispeech/speechocean762")