I'm curious how the economics work out with building your own TTS vs. just paying Google to do it for you. Of course Google's TTS API costs money, but I assume that you also had to pay for a dataset of transcribed speech. After how many hours of audio is the breakeven point?