Question about Recording Voicebank

Well, after I'd put together my UTAUloids voicebank, I was poking around different tutorials on youtube out of curiosity. Someone in one of them mentioned something interesting that I considered trying, but wanted to check the validity of it before I go off recording another 100+ files. :I [My throat is still sore from the first].

Anyway:

Is it true that if you record longer notes for each syllable ["A", "Ka", "Ga", etc.] the more humanesque the UTAUloid will sound? Never completely human, but they wouldn't sound like a complete robot either?

[Or is such a thing only effected by the oto.ini?]

Thank you for your time.

i've noticed by experience that this is true...

to a /very/ small degree.

it mostly depends on the recordings, a 1 second recording compared to a 3 second one only sounds /slightly/ more robotic... barely noticable with background music and whatnot (and mostly on long notes, like those in Can't I Even Dream).

when i originally started, the common advice was long recordings so all of mine (Jessi Maki) are 3 seconds and over, if you want to hear an example of longer ones.

Let's say a normal sample with its already oto is like this:


Well, the longer the white zone, the vowel part, the more human the very long notes will sound.

UTAU has to stretch the vowel if the sample is shorter than the needed length of the note (for example, if the sample has a vowel of one second and you want to make a note of two seconds UTAU has to make that vowel part longer artificially stretching the sample).

But, of course, this only applies for really long notes. Probably samples with vowel around 0.5-2 seconds are more than enough but vowels of like 0.2 or less may start to be a noticeable problem.
As an example, the cute robotic Uta Utane has a vowel around 0.1 seconds while the very cool "new" Aoi Celestine sounds very human with vowels just a bit over 0.5 seconds -but Defoko may be a bad example because she doesn't come from a person from the start-.

If you plan on doing long samples for that, look again to the picture of the oto config of the start.
You see a red line going around the center up and down (not the vertical line of the oto config). Well, if you make the sample long but the line goes away a lot from the center, then don't record them that way. That frequency thing part (not sure what exatly is!) is more important that the sample being long.


If you want to sound more human, try the "blanket over your recording" trick. It really helps a lot to improve the recording.

Well, admittedly, I'm not too concerned with her sounding 100% human. If I wanted that, I wouldn't be messing with UTAU right? xD I'd be recording straight. But UTAU is fun, and I like the slight robotic quality to their voices.

I suppose the main thing I'm concerned about is that in some songs, some notes are really...awkward? That might be the good word.

For Example: I made this for my UTAUloids first song, as a test run of sorts.

https://www.youtube.com/watch?v=srXhYSwFluw

There are various parts of the song [Most notably the "Talking parts" and about 2:32-2:35 (and some others, but it's most prominent there so you know what I'm talking about)] she doesn't..pronounce the notes right. Thing is, I haven't the first clue about USTs, so I don't know if the notes are set up awkwardly in there, or if it's my UTAUloid and her configuration or something.

It's just a strange thing, and I'm not sure who or what is at fault yet so I can go about fixing it. As you can see, she sounds fine on most other notes. :<


Thank you!

aaa, ire1as always has pictures xD so much more helpful~~

but, to put my opinion in, it sounds like an OTO issue to me

or the .ust needs to be fitted to her voice..? sometimes certain VB's (configured properly or not) won't work right with premade .usts

Alright. Thank you! I know her oto file needs a looot of work. :I So. It's probably that. Derp.

Thanks again! You guys are so awesome~

I agree with Kolya-kun that it could be the oto or the ust not being configured for your voicebank.
Check this awesome post as I think it may be a problem with the ust (I'm assuming your oto.ini is well done):
https://overseasutau.forumotion.net/tutorial-and-resources-directory-f31/fitting-a-ust-to-an-utauloid-t2935.htm
(I'm spamming around this post!... uh, don't ban me, please.)

About the talking part... it's going to sound bad >.> Doing talkloid (or hanasu or whatever it's called) is hard for UTAU, ugh. The pitch needs to be configured almost each pixel with mode 1, ouch.
Very good results are doable but it really needs quite a bit of work (oh, well, there is a great plug in that helps a lot but you still need work) and practice with pitch checking programs.

Thanks for the link! I'll definitely check all this out. My Oto isn't..uh. Well, it -is- done. But not by me. :\ So I've not the foggiest clue if it could be considered "well done" or not.

But thank you again! [And thanks for the sub on youtube! x3 ]