7. VCV Japanese
I know you're probably thinking "Wait, I thought CV was easier!!! Why are we starting with VCV??"
Well, the theory used in CV and CVVC banks relies heavily on the theory used in VCV otos. Otoing relies heavily on the relationship between a syllable and the vowel before it, and this relationship is most easily seen in a VCV oto.
Well, the theory used in CV and CVVC banks relies heavily on the theory used in VCV otos. Otoing relies heavily on the relationship between a syllable and the vowel before it, and this relationship is most easily seen in a VCV oto.
The Stats
Average recordings: ~300
Average size: ~1000 lines
Typical Aliasing Method: Hiragana [a か]
Average size: ~1000 lines
Typical Aliasing Method: Hiragana [a か]
The Mechanics
A VCV oto captures one CV pair and the vowel in the recording before it - you may have seen it written "V-CV" and that's why. VCV is considered by most to be the smoothest sounding voicebank type, and that's because it fully captures the natural fade-out between vowel and consonant. This is easiest to see when we open an oto segment in UTAU's editor (see Figure 1).
As you can see, the vowel distorts a bit before the consonant, and this distortion is unique to each consonant as well. A VCV oto captures this transtion perfectly, with the exact timing of the recording.
The Base Oto
Now unless you wanna sit there for three hours splitting 300 recordings into seven lines each and then naming them, I suggest you use an auto-generated base oto, even if only for the aliases. There are two programs that offer auto-generated base otos, and they each have their own pros and cons.
Moresampler
|
Setparam
|
Choosing Base Parameters
You may have noticed that for setparam's base oto you need to input values for the four main parameters. Even if you used moresampler, it's a good idea to change the parameters it gave you to your own base set with copy and paste. But... what do you set it to?
It's a harder question than you think. VCV otos depend heavily on the tempo you recorded at: faster tempos need lower values and vise versa. I always record at 120 BPM, and set my values like this:
Overlap: 100
Preutterance: 250
Consonant: 300
Right Blank: -700
Note the negative right blank value - this causes the right blank to be measured relative to the left blank like everything else, instead of the end of the recording. It also allows it to move when we move the preutterance. That'll be useful later, when we're actually otoing.
If you generated your oto with moresampler, you'll want to copy-paste your custom values over the ones it gave you. To do this, either fill out the first row with your preferred numbers and select all four to copy it, or write in notepad something like "100,250,300,-700" and copy it. Select the top row, where the names of he parameters are, in order to select the whole columns, and then paste. You may get a warning about the sizes being different; just hit "yes" and wait for it to finish. Before you do anything, save your oto, close setparam, and open it again. This will keep setparam from changing your negative right blank values to positive ones.
One more thing - VCV reclists are not very efficient, which means you'll end up with 50-100 duplicate aliases. Fortunately there's an easy way to get rid of them without having to sift through the whole oto, as described in the last tutorial.
Overlap: 100
Preutterance: 250
Consonant: 300
Right Blank: -700
Note the negative right blank value - this causes the right blank to be measured relative to the left blank like everything else, instead of the end of the recording. It also allows it to move when we move the preutterance. That'll be useful later, when we're actually otoing.
If you generated your oto with moresampler, you'll want to copy-paste your custom values over the ones it gave you. To do this, either fill out the first row with your preferred numbers and select all four to copy it, or write in notepad something like "100,250,300,-700" and copy it. Select the top row, where the names of he parameters are, in order to select the whole columns, and then paste. You may get a warning about the sizes being different; just hit "yes" and wait for it to finish. Before you do anything, save your oto, close setparam, and open it again. This will keep setparam from changing your negative right blank values to positive ones.
One more thing - VCV reclists are not very efficient, which means you'll end up with 50-100 duplicate aliases. Fortunately there's an easy way to get rid of them without having to sift through the whole oto, as described in the last tutorial.
Time for the Fun Part
It's time to rock'n'roll. Turn on some tunes and let's get started.
The first value we're going to set is the overlap. Do not change the numerical value of the overlap. Instead, slide the left blank or preutterance until the overlap is in place. The overlap should be placed at the end of the consistent part of the vowel, before the vowel fades or changes pitch. See figure 3.
The first value we're going to set is the overlap. Do not change the numerical value of the overlap. Instead, slide the left blank or preutterance until the overlap is in place. The overlap should be placed at the end of the consistent part of the vowel, before the vowel fades or changes pitch. See figure 3.
Then, change the value of the preutterance so that it's after the next consonant, before the vowel (see Figure 4). For sounds like "kyu," include the y in the consonant so it doesn't get lost in the overlap with the next sound. Don't make it too small though; if the preutterance seems too big move it behind the consonant and don't worry too much about the overlap. The overlap is important, sure, but the preutterance determines the timing - meaning it takes priority.
It's also worth noting that the preutterance shouldn't be too big either - you don't want it overrunning the previous vowel in faster songs. I tend to limit myself to four times the overlap, but even that seems like a bit much at times. |
The consonant and right blank are easy - drag them until the area between them is the most consistent portion of the vowel for both pitch and waveform. The vowel's waveform should be really easy to see on the spectrum - just look for the part with the most consistent pattern. Then just move on to the next syllable and repeat.
Exceptions?
VV pairs are a little bit different - and thank god for that spectrum am I right? Look at Figure 5, if you were just using the wave form you'd never be able to spot the difference! What you wanna do with these is treat the transition between the vowels the way you'd treat a consonant - make sure the overlap is before the first vowel becomes inconsistent, and the preutterance where the second vowel is clearly itself.
- CV's are easy; since there's no previous vowel to worry about, just put the preutterance after the consonant and you're gold. The method for setting the consonant and right blank is always the same. |
This all probably seems intimidating but with setparam it'll be over before you know it! On to the next bank type.