Just the recognition part will tailor to the local version of Mandarin, which is why they need to seed the teams with thousands of hours of local corpora. Many of the teams, they’re using existing speech sets, and it’s not really representative of the local way that people speak Mandarin.