I tried to build something like this a couple of years ago, just by using speech-to-text, and that worked really badly (but it was better than nothing). I've given BoldVoice a spin and am very impressed by how well it works on phoneme level - well done, congrats on launching!