Then, they crowdsource this into the deep speech engine so that they can recognize the language as is spoken in Taiwan instead of forcing, like in Siri, everybody to speak a perfect Mandarin in order for the machine learning to recognize. I think this is also very creative.