I ever learn how people make UTAU (something that similar to Vocaloid) voice bank. Rather pick someone else voice (that not trained and binded to a certain language), why not make it versatile? I mean at current AI voice, the laugh is not good and the high pitch while talking is making the voice gone too. With make wide range voice sample, it would make it clearer. I think the problem is on the sample voice and the combination vocal vowels that it made. If the developer could, maybe, you could use some of that UTAU trained voice. Their tons of it and sound goods too. I suggest take a look at Denatsu Sora or Rinne Tsubasa.
