Here is an interesting article about Vocaloid 2, a program developed by Yamaha that, given lyrics and a melody, generates anime-style singing.
The singing sounds like a normal human voice modified by digital effects, such as one can hear in some of the music kids listen to these days (and I am assuming here that those songs do not use synthesized singing). Apparently the software is a big hit in Japan (go go user-generated content – the Innovator’s Dilemma at work again).
I wonder if singing is easier to synthesize than speech. I wonder if this is related to the fact that accents are harder to make out in singing than in speech. Further restraining the problem domain to the kind of singing one hears in anime-style songs probably makes things even easier.
Still, a promising step on the way to usable text-to-speech. I can easily imagine an anime-style game using this technology.
(Via Boing Boing.)