Details Model trained with train_size=0.2 or 20% of the Mozilla Common Voice Arabic training+validation dataset.