Hi, I just managed to install and run Deepspeech to transcribe an hour-long interview with an official in the Western Cape, South Africa. The accent is strong, and the inference output made for some pretty wild reading. So much so that I think it might be quicker to ask a student to do it manually than to read through this and fix it…so… are there models based on South African or Western Cape English that will produce something more sensible? Or is there a way to train the programme to transcribe more accurately?
Do you have South African / Western Cape English data?
If you mean a database with a model of Western Cape speech, I don’t have one. I may be able to source something from relevant linguistics departments at local universities, but wouldn’t know who to approach.