Building Cost-Efficient, Robust and Linguistically Diverse Automatic Speech Recognition Systems in Pan-African Clinical Context