8. CV Japanese
Now that we understand how VCV works, CV's a walk in the park. CV banks tend to sound choppy because the natural vowel fade out isn't in the oto at all - but with a little effort we can make it sound just as smooth! Or at least, iron out most of the kinks XD
The Stats
Average recordings: ~150
Average size: ~155 lines
Typical Aliasing Method: Whatever's opposite of the encoding method
Average size: ~155 lines
Typical Aliasing Method: Whatever's opposite of the encoding method
The Mechanics
The trick to a good CV oto is to put the overlap where the end of the vowel would be if it was a VCV bank. That way, when you hit p2p3 to set the envelopes, the crossfade will mimic the natural vowel transition.
The Base Oto - Choosing Base Values
The thing is CV otos are tiny - only a tenth the size of a VCV oto. And, since the recordings don't need to be split, there's really no need for an auto-generated base.
Still, there is a bit of setup involved. You should alias your recordings in the opposite method of how their recorded. I.e. if a bank's file names are all hiragana, you should use romaji aliases, or vise versa. If you make a Japanese bank that can't be used with Japanese characters, you need to rethink your life.
In addition, if the vowels were recorded separately instead of in VCV style strings, you need to duplicate the recordings and add a couple aliases. For example, "a.wav" would need to be aliased [あ], [*a] and [*あ]. You'll see why here in a second.
The only other thing you need to do to set up the base oto is paste in base values. I tend to use 60,90,120 and then set the right blank to something huge like -800 or -1000. Don't worry about the left blank, it's not hard to just scroll over to where the sound starts.
Still, there is a bit of setup involved. You should alias your recordings in the opposite method of how their recorded. I.e. if a bank's file names are all hiragana, you should use romaji aliases, or vise versa. If you make a Japanese bank that can't be used with Japanese characters, you need to rethink your life.
In addition, if the vowels were recorded separately instead of in VCV style strings, you need to duplicate the recordings and add a couple aliases. For example, "a.wav" would need to be aliased [あ], [*a] and [*あ]. You'll see why here in a second.
The only other thing you need to do to set up the base oto is paste in base values. I tend to use 60,90,120 and then set the right blank to something huge like -800 or -1000. Don't worry about the left blank, it's not hard to just scroll over to where the sound starts.
Git 'er dun
Like before, you shouldn't change the numerical value of the overlap - just adjust the preutterance or left blank until it ends up where it needs to be. Like I said earlier, he key here is to put the overlap where the previous vowel would end in a VCV recording. The preutterance is the same as VCV - behind the consonant, beginning of the vowel. Note: Unlike VCV, sounds like "kyu" should not include the y with the consonant. The consonant and right blank methods don't change. They should capture the consistent part of the vowel. One thing to note is that that region should be much, much larger. At least a whole note at 120 BPM. That way the quality when the note is made larger doesn't drop.
I mean, it's CV, you gotta take all the quality you can get.
I mean, it's CV, you gotta take all the quality you can get.
What about the vowels again?
Right, so if you didn't record your vowels in strings you actually need to oto them twice: once for when they begin a string and once for blending in the middle of the string.
For example:
"a.wav" is aliased [あ]. That line should have the preutterance at the start of the vowel like normal. But it also has line [*a] and [*あ]. Those lines should be identical, and are used for blending. In order to simulate the natural vowel blend, set the overlap to twice the preutterance (I usually use 120,60) and make sure the left blank cuts off the beginning of the vowel entirely. Then copy the parameters from the line you otoed into the alternately aliased line. That way when you hit p2p3, the crossfade will mimic the natural vowel blend almost seamlessly.
Well that was easy. Time for the last common Japanese bank type: CVVC
For example:
"a.wav" is aliased [あ]. That line should have the preutterance at the start of the vowel like normal. But it also has line [*a] and [*あ]. Those lines should be identical, and are used for blending. In order to simulate the natural vowel blend, set the overlap to twice the preutterance (I usually use 120,60) and make sure the left blank cuts off the beginning of the vowel entirely. Then copy the parameters from the line you otoed into the alternately aliased line. That way when you hit p2p3, the crossfade will mimic the natural vowel blend almost seamlessly.
Well that was easy. Time for the last common Japanese bank type: CVVC