it's Acapela. it's impossible to build your own text to speech engine if you want quality - it's years of R&D to have a very good one.