Is the approach being used to do accented TTS (or just reference recordings), and then a tone color conversion model that just changes the timbre? Because if I say a completely different sentence it still says the original words, haha.