I'm curious, what on-device text-to-speech engine did you use?