coraal
Audio recordings and transcripts of regional African American English varieties for ASR.
Open dataset Updated 2026-06-18
What is coraal?
CORAAL is a collection of audio files paired with orthographic transcripts documenting regional varieties of African American Language.
The dataset is intended for training and evaluating speech recognition systems on non-mainstream English dialects.
What you can build with coraal
Train dialect-specific ASR models
Fine-tune Whisper or Wav2Vec2 on the time-aligned transcripts to improve recognition accuracy for African American Vernacular English across regions.
Build inclusive voice applications
Develop speech-to-text features for apps that better handle regional AAE variations in customer service or accessibility tools.
Conduct sociolinguistic research
Analyze phonetic and lexical patterns from multi-region recordings paired with aligned text for academic studies on language variation.
Load coraal
Python
from datasets import load_dataset
ds = load_dataset("bezzam/coraal")- 1pip install datasets soundfile
- 2from datasets import load_dataset
- 3ds = load_dataset('bezzam/coraal', 'default')
- 4Run included preparation scripts to convert to ASR-ready format
- 5Access audio via ds['train'][0]['audio'] and transcripts via ds['train'][0]['text']
coraal: pros & cons
Pros
- +Multi-region African American speech coverage
- +Time-aligned transcripts ready for ASR
- +Preparation scripts simplify preprocessing
- +Direct Hugging Face datasets loading
Cons
- –Focused only on one demographic group
- –Fixed 2023.06 release version
- –Requires external library for audio handling
Did you find this helpful?
Frequently asked questions
CORAAL provides speech recordings from African American speakers in multiple U.S. regions with time-aligned transcripts, formatted for ASR via preparation scripts.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar datasets
Other audio & speech options worth comparing.