Dataset · intronhealth
200 hours of English speech with African accents
intronhealth/afrispeech-200
An audio corpus of English speech with the accents of 120 African peoples, including medical vocabulary, for ASR tasks.
What's inside
The dataset is a 200-hour English speech corpus (from 10 to 100 thousand recordings) recorded with the participation of 2,463 unique speakers from 13 African countries. The material covers 120 African accents and includes both general and clinical (medical) subjects. The project was created by Intron Health to develop speech recognition systems in African regions.
What you can do with it
- Fine-tuning ASR models (e.g., Whisper) for robust recognition of various African accents
- Developing voice assistants and dictation systems for the medical field and clinics
- Evaluating the quality and fairness of existing English speech recognition systems
- Tasks
- Automatic speech recognition
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 3.4K
- Likes
- 45
- Access
- on request on Hugging Face
Models trained on it
- ogunlao/SBPN_multilingual_baseAutomatic speech recognition 171
- ogunlao/SBPN_multilingual_largeAutomatic speech recognition 68
- DivineAngel317/whisper-small-nigerian-accented-englishAutomatic speech recognition 57
- Qlerqly/griot-nano-1Automatic speech recognition 28
- Professor/whisper-small-afrispeechAutomatic speech recognition 19
- samwell/medasr-ghanaAutomatic speech recognition 12
- Professor/whisper-medium-afrispeechAutomatic speech recognition 11
- vedaradhak/bac-accent-transcriberAutomatic speech recognition 3
How to load it
from datasets import load_dataset
ds = load_dataset("intronhealth/afrispeech-200")