I recorded six minutes of English singing data on my phone and turned it into a Diffsinger model :)