1st tweet (Ryo worked on a prototype for more real-time voice analysis):
*FFT = Fast Fourier Transform. It's done in real-time, it's confusing but it's basically an algorithmn that analysizes (computes) a signal from its original domain (whatever that means) into a representation.During summer holiday, I implemented a newly thought-up real-time vocal analysis synthesis system.
Because it implemented signal processing everything except the FFT* from scratch, it was quite difficult, but this is the prototype.
However, the input voice in the movie was borrowed from "Seiyuu Toukei Corpus".**
**Seiyuu Toukei Corpus = 日本声優統計学会 It's a site for "voice actor statistics", it has patterns of 3 voiceactresses and can be used by people for machine learning.
2nd tweet (real-time voice conversion for iPhones):
I'm not gonna translate this word for word, but Ryo says that for the algorithm, but he's working on controlling the increase in processing time and that they are creating an iPhone version.
My thoughts:
Seems like might be related to improving Cherry Pie? Beef Jerky uses data from voice actresses (I assume in order to make the converted speech talk naturally).
Ryo likes naming things after food. Wonder why it's called Beef Jerky.
If this iPhone version comes out, it really does seem similar to R.C. Voice vs Voidol (Crimson Technology's real-time voice converter for iPhone and computer respectively). Crypton must be trying to be competitive. Doesn't this kind of prove that Cherry Pie will not be Vocaloid-related if it works in this manner?