Microsoft’s text-to-speech model VALL-E can mimic anyone’s voice with little training data. Now it can do so across languages while replicating emotion.