Hm. I don't know what else they're inputting besides vibrato - it didn't seem to change its vowels when it changed pitch, for example, or slide between notes. Maybe they do stress/volume information separately rather than the computer being able to do it?

For speech, IBM's Watson was pretty good.