Aster (stella azurea)

More info coming soon.

I don't mean this negatively, by the way. I think it's good for vocalsynth to maintain at least *some* degree of artificiality/uncanniness that distinguishes them from actual human singers.

For those who have ever asked "I can't tell the difference between vocalsynths and real singers anymore," here's a tip:

Even the highest-fidelity vocalsynths struggle with clear pronunciation on particularly high or low notes. I personally think it's the easiest telltale sign.

The project I've been hoping to announce has been taking longer than expected, but I hope to have something to show by the end of the year. Please stay tuned~

The even worse part is that for vocalsynth, the use of "AI" here actually predates the current fad (it started around 5 years ago to illustrate its use of machine learning compared to traditional models). There's a huge feeling of "we were here first!!” while we get stigmatized 🥲

Because I'm sick and tired of people assuming Vocaloid/CeVIO/SynthV "AI" is generative AI just because it has the word "AI" in it: a quick and easy explanation on what vocalsynth "AI" is and why you need to stop freaking out.

From what I hear it doesn't show up with the standalone version, so maybe try it with the installer?

A screenshot of AviUtl ExEdit2 with an English interface.

What do you mean AviUtl 2 has *official English support* spring-fragrance.mints.ne.jp/aviutl/

Someone did anti-AI glaze for audio!

in particular, the incident where a single patch resulted in an algorithm trained too much on one VB making the others sound awful is particularly illuminating, because it demonstrates it's really not useful outside a certain limited context

It does use some of the algorithms used in generative AI, but I'd still classify it as procedural in terms of overall usage because it largely has to affect an existing base (the notes you input) for it to function properly

Basically exactly like that!

I really need people to understand that the only thing AI does here is "make it not sound uncanny". You still have to write in the notes manually. Still have to do tweak pronunciation and pitch manually. Still have to post-process manually.

Hey, uh, just so people know: the type of "AI" primarily used in the context of vocalsynth is not generative. It's procedural. It's to induce randomness in ways like "making water ripples realistic" is done in games.

You still have to do everything else by hand.

Painstakingly.

Notice regarding change of handle name and branding stella-azurea.com/notice-regar...

Happy to see so many people seem to have remembered to find me even without me announcing it 🙏 Will try to post more on here!

Sounds good to me!

I say 1 or 3, 1 invokes the idea "Aiko from within the Kikyuunes" and 3 invokes how I'd search her name

'sup

...wait a second.

Quite a few people found me here despite me not having made an announcement (yet), just by name alone. Thank you for remembering me 😊