How do VCV voicebanks work?

[StaticRadio]

I've watched a couple of video's on how to make VCV voicebanks but they've been no help at all. I know that you have to record syllable strings (Like ba ba be ba bi ba bo ect) and that in theory, when you oto them you keep the vowel from the previous note (Like the a from ba going through to be) so it would be a-be and the note would be smoother. But I don't get how you name them and oto them, or really how you record them? Do you record the syllable string and select the part you want to export into Utau? And how do you name them? It seems like you would have to record a ton of sounds, because to get them to be smooth you would have to record all sounds with be, all sounds with bo, and so on. [sub]Like be ba be ka be cha and go through the entire alphabet with be, so to speak
I probably sound like an idiot, but I'm just really really confused orz

 

[lapinblanc]

[/watches thread intently]

I have no idea how to make a VCV bank either, pal. I'm hoping I can learn something here too. LOL

 

[Halo]

Download, use and study a VCV Voicebank.
It'll help a lot.

As for the theory behind the method, You record strings and then duplicate the oto for that file and alias it accordingly.
So, if you just had one string, aekakesa, the oto would look like
aekakesa.wav= "- あ",0,0,0,0,0
aekakesa.wav= "a え",0,0,0,0,0
aekakesa.wav= "e か",0,0,0,0,0
aekakesa.wav= "a け",0,0,0,0,0
aekakesa.wav= "e さ",0,0,0,0,0

This prevents you having to split up the file manually.

There are plenty of premade recording lists, from 2-syllable strings (_まか) to 8-syllable strings (ああいあうえあ). My favorite is the full 5-syllable list that Haruka Nana and Hibiki Shinji use.
Also, in the end all of these lists have the same amount of syllable combinations, so just because one list has more syllables per string does not mean you're doing more recording.

For a full list there are about 1,200 syllable combinations all together, but there's only roughly 300 recordings.

But, again, it makes more sense if you actually work with an existing bank.

 

[lapinblanc]

How do you go about adding aliases? From all the combinations, I'd imagine that would be a lot of work by itself! As well as the oto.ini ... Are you supposed to edit every sample one-by-one? Or is there another way? I feel the oto.ini process scares people the most (me included)!  If I were to use Haruka Nana's VB, is there a text reclist for her I can upload on, say, OREMO, to make recording easier?

LOTS OF QUESTIONS, HNNNNGGGG

 

[Halo]

You know how OREMO has reclists you can load?
It has a function to make Reclists out of the WAV files in a folder... AKA a voice bank. So you can just point it to Nanas VCV folder and click "make a reclist out of this folder" and tada!

Otherwise, there are VCV reclists that people hand-transcribe all around the place.

Now, if you record in hiragana the oto is no problem at all. Setparam has an auto-otoing function that'll automatically duplicate and alias for you, halving the amount of work (you'll still have to go and check every file to make sure it's oto'd correctly, but still).

If you want to oto by hand or record in romaji but don't want the pain of aliasing, you can just take someone else's oto (from the same list) and clear everything but the aliases in setparam, and it should work just fine in your bank.

By the way, when you're recording, do yourself a favour and use a Guide BGM or metronome. This makes otoing so much easier in the end.

 

[lapinblanc]

You know how OREMO has reclists you can load?
It has a function to make Reclists out of the WAV files in a folder... AKA a voice bank. So you can just point it to Nanas VCV folder and click \"make a reclist out of this folder\" and tada!

I did not know this... Thanks for clarifying that!! I also didn't know what Setparam did ... thanks for giving me an idea! :D

 

[Halo]

Oh, to clarify, SetPeram is an otoing tool. Makes otoing a lot faster once you learn it, but otoing by hand in it for VCV oto's results in kind dodgy otos-- So it has a VCV auto oto tool included.

It works by inputting what tempo you recorded at + about how much space you left in front of each sample, and from that it guesses everything (and aliases for you). Unfortunately, it only works in hiragana. u vu;