Other M. Morise vocal synth project releases Tohoku Kiritan's singing data

[uncreepy]

Back in September (around the announcement of Next Generation Piapro), there was a Tweet from Zunko about a project M. Morise was working on relating to Kiritan. Here is the old translation from the General Discussion thread:

I'm excited by the talk of Miku-san switching engines, but ヽ(*'▽'*)ﾉ I'm a Vocaloid, Utau, Voiceroid, and even Voidol; and while Kiritan is preparing to go public with research data with M. Morise of Meiji University to be on a new singing synthesis engine she's in a state of chaos ♬♬٩(๑❛▽❛๑)۶♡

That was really messy/hard to translate in proper English. Zunko is reminding us all of the engines she's on. And then she teases that Kiritan will have a singing voice (she currently has one for UTAU) for a new engine worked on by M. Morise.

M. Morise's full name is Masanori Morise and he is a researcher who worked on STRAIGHT and WORLD. To my knowledge, STRAIGHT is a speech analysis and synthesis framework that can be used by the public (not commercial use)

WORLD is superior to STRAIGHT (uses less computing power) for real-time singing synthesis that is supposed to be as natural-sounding as the input speech. It is a vocoder. CeVIO and UTAU both use WORLD for their engines.

M. Morise's Twitter: M. Morise (忍者系研究者) (@m_morise) | Twitter

I assume the research is to turn a version of WORLD into a commercial real-time voice converter?

New information:

Kiritan's singing database has been publicly distributed. In order to download it, you have to click the blue button under the image of Kiritan on website that says "Facebookでログイン (log in with Facebook)". It contains 50 songs which are a cappella.

There is a thread in this tweet containing more information. The data is public because Japanese copyright law changed, letting people view this data so that people can conduct their own research (not just for making a Vocaloid-type synthesizer, but to contribute to the research of vocal synthesis by researchers). It's related to a new business purpose project by M. Morise.

The .zip file contains:
midi_label (music score info)
mono_label (phoneme label info)
wav (sound clip of the a cappella singing) [totals to almost an hour of singing]
japanese.table (dictionary file)
list [Excel spreadsheet] (list of songs)
noise [text file] info about noise in the recorded voice
readme

If you actually make a vocal synth out of this, you have to get permission from SSS Co., Ltd.

 

[uncreepy]

Kanru Hua (Synthesizer V creator) tried his hand at making Kiritan sing "Bad Apple!!":

It would be cool if Kiritan got to be on Synth V. (Would be surprised if it happened for real, though.)

 

[Prism]

Low key sounds better than 90% percent of vocal synths.