Browse Datasets

Med-Convo-Nig

Simulated Nigerian-Accented Medical Consultations

asr speech-recognition medical-transcription 4.2 hours audio CC BY-NC-SA 4.0
Access Dataset

Description

Simulated doctor-patient tele-consultations recorded in Nigerian-accented English (about 4.2 hours of audio with transcripts), released by Intron Health. Gated on Hugging Face: request access before loading.

Quick Start

Python 
from datasets import load_dataset

# Requires access approval on Hugging Face
dataset = load_dataset("intronhealth/med-convo-nig", split="train")

Project Ideas

Medical Whisper

Fine-tune Whisper on Nigerian medical terminology to improve transcription accuracy.

Intermediate

Speaker Diarization

Build a pipeline that separates doctor and patient speech automatically.

Intermediate

Ethical Considerations

  • Simulated data limits patient-privacy risk, but discuss ASR bias against regional accents in real care.