Hi, good question: the main purpose of the challenge is to collect a brand new dataset in an African language. The model training part of the submission is just for you to demonstrate that you've learned to train an ASR model (the model is more of a completion task - since everyone's data will be different, judges won't look at how well your model performs cause you can't do a comparison). The judging will be based on the dataset you created (which can't be an existing open source dataset). Hope that answers your question.