Capitalising on North American speech resources for the development of a South African English large vocabulary speech recognition system

Herman Kamper, Febe de Wet, Thomas Hain, Thomas Niesler

Abstract: Highlights•We try to develop or improve a South African (SA) speech recogniser with US data.•The SA domain is under-resourced while the US domain is very well-resourced.•US pronunciations and language models can feasibly replace SA counterparts.•US acoustic data used in a SA system results in a large performance penalty.•US acoustic and language model data slightly improve a SA system by adaptation.