Wiki Wednesday #313 - VoiceFace's UFO-Chii / UFO-地異

 I'm so tired all of the sudden.

Wiki Wednesday #313 - VoiceFace's UFO-Chii / UFO-地異

Ufo-chii is a comedian from another solar system. They get upset when none of their jokes land. Earth people just can't get the jokes!

Who Is UFO-Chii?

Art from site

.

Did you know that there's an error in the DIFF-JA phonemizer? And that Synth V has the same error? 

Most people would have no idea what I'm talking about. When I explained it to a Japanese person, she had no idea what I was talking about either! 

In Japanese, with a few exceptions like "si" or "di" in loan words, all consonants followed by "i" are palatalized. What does that mean? You know the "n" with the silly hat in "Niño"? That "ñ" is palatalized! Italian has that phoneme also - the "gn" in "lasagna". Now, as an American that just sounds like an "n" followed by a "y"! That isn't the case! 

I'll cut to the chase. Most people will only be labeling data sets using their own voice or the voices of their friends. If you think that ひ (hi) is pronounced as the English word "he", then you don't need to worry at all. The phonemizer works perfectly for your banks and you can skip to the part where we see how this utau sounds.

If you think ひ is most closely pronounced like the English word "she", this applies to you.

This is an easy fix as long as you use DIFFS for the phonemizer. Edit the dsdict so that the correct phonemes are used. So why do I care enough to label my data sets "incorrectly" and add an extra step for optiomal usage? (note: I probably made mistakes, but you should be able to create a disdict_ja to fix this issue.)

I have autism. Just literally that. I identify too much with technology and I put myself in its shoes. If I was given a data set and I was meant to learn each phoneme, I would be upset if sometimes I expected "k" but instead I was given "ky". Labeling it "incorrectly" in the fandom's view results in one issue - stuff like "ki" and "mi" sound a little funny when using the default Japanese phonemizer or an unedited dsdict. It has no idea what [k i] or [m i] sounds like because that does not exist in the data set. If you label them "correctly" so that it has [k i] in the data set, almost all non-palatalized consonants will have incorrect data. This really isn't a huge deal for the most part, and I totally get that. I have only been able to train myself to hear the difference for a few consonants. But when I can tell the difference... It's a huge difference! The idea of mixing up "h" and "hy" bothers me so much, I have to fix it all! 

Besides, if you don't do it correctly, you're almost never going to run into "by" or "py" organically! There's been two times I've gotten Japanese data sets and there have been no "py" phonemes. One of them just recorded a line from an utau recording list because they couldn't find any lyrics... Because they didn't know that ぴ and ぴょ used the same phoneme. This proves this isn't just a Western issue! It's an everyone issue! 

How Is UFO-Chii's bank?

Ufo-chii has a two pitch CVVC bank. They have a very nice deep voice!

Where Can I download UFO-Chii?

You can find them on their official site! This is a cool UTAU!