Dataset Details
Title:
MedASR-Ghana: Medical ASR for Ghanaian-Accented English
Details:
This model is a fine-tuned version of Google's MedASR optimized for Ghanaian-accented English speech recognition, particularly suited for clinical and medical transcription in Ghana.
Model Description
MedASR-Ghana is designed to transcribe English speech from speakers with Ghanaian accents, including Twi, Akan, and Fante language backgrounds. It builds on Google's MedASR foundation (a 105M parameter Conformer-based CTC model) and adapts it specifically for West African English pronunciation patterns.
Key Features
Optimized for Ghanaian accents: Trained on Twi, Akan, and Fante accented English
Medical domain ready: Inherits MedASR's medical vocabulary capabilities
Lightweight: 105M parameters - efficient for deployment
CTC-based: Simple greedy decoding, no language model