https://youtube.com/watch?v=9Mm1TIG0A_k [Embed]

The presentation is here for those who want it. It should be the very last one in this stream archive. The specific section is about 5 hours and 15 minutes in.

That's fascinating. It's clear that Crypton has a vision for voice synthesis that's very different from the SV-type direction. SV is basically iterating on traditional Vocaloid workflows with a more detailed timbre, and here NT is rethinking it with a simpler timbre.
I appreciate the unique perspective, it's good for the scene. All Crypton has to do now is make NT sounds as good as, like, vocaloid2, lol.

>>27707
I think they do sound pretty V2-ish now compared to before but regardless there’s definitely still room for improvement, especially with the loud engine noise. With that in mind I’m curious how it’s going to all work out. I remember Noboru (Internet co’s pres) expressing interest in the past. I imagine a Gumi themed editor lol

>>27705
Does that imply the possibility of non-crypton NT banks?

>>27726
Not just that, but think of potential third parties also bundling in their own editors. For example if we got GUMI NT but she came with a Megpoid Editor. It’s still NT2, but you can make your own creative application with the engine.

Tei NT... I want to believe...

>>27753
Can't wait for Hatsune Gumi, Hatsune IA and Hatsune Flower

>>27755
That's SP not NT

>>27756
NT sounds as bad if not worse. Or at least the non Miku banks do.

>>27767
Nah, NT actually sounds better now ngl. Cfgjndh Engine noise aside, there’s been small tweaks a lot of people haven’t really noticed, more specifically with the transitions between notes. SP is really choppy, current version NT2 is smooth. I should probably make another render for reference but I have a cover on my second channel I made with an old file during the NT2 beta. Loading the file now, gets different results. There’s not a lot of good examples of it admittedly, but that’s quite literally how I am able to tell the difference between SP and NT now lol

I do have concerns to some others with banks sounding “mikufied” as well but I think we might be safe if it’s a potential third party bank.

Non-crypton NTs would be a good thing now that i think about it because when was the last time we got a voicebank that wasnt AI

>>27778
Aside from Rei sometime this month, I think it was SP. I’m not sure if NT2 is necessarily the same synthesis method? It seems like it’s a hybrid of waveform and AI now? I’m not sure though. Someone can probably explain it better than me.

>>27781
I don't think NT is AI(?)

>>27778
Without counting Rei who hasn't been released yet there was SP and Rin/Len ST.

>>27789
If you watch the whole presentation they explain the Tech and M9 engine. It does use neural networking for the processing but the banks themselves still use samples and stuff. Like I said, it seems like it’s a hybrid now. NT1 was Spectral Resynthesis for reference. It could still be Resynthesis but use AI now but it’s funky. This is the most information we have on how M9 works in NT2 right now but I can’t really explain it or properly understand it.

>>27790
Didn’t count Rin/Len NT because I’m not completely sure about NT2’s engine rn be it looks like a hybrid of some kind.

IMG_9961.png - 457.77 KB (1334x750)

IMG_9962.png - 454.34 KB (1334x750)

IMG_9964.png - 455.11 KB (1334x750)

Can’t count anything NT1 at the very least because it’s spectral and not Concat technically. Similar to V1. I don’t know how different NT2 is by comparison but there is AI involved. I think I might be missing a slide but the AI tmk at the very least it utilized in not only the Automatic tuning function (which really doesn’t do a whole lot) and the Vocoder specifically. The slide that mentions the process is what confuses me most. It goes through a process in the M9 engine then the waveform output, waveform being concat.

I don’t have an audio transcription for anything prior to the NTSDK because that’s what personally interests me most, so someone with a bit more understanding of the language might want to explain it better if the watch the whole thing, but yeah, it’s definitely interesting I’d love more clarification.

>>27797
Even if it's a hybrid it sounds like the sample based voicebanks so i'm cool with third party NTs. Hope we get Kaai Yuki or Gakupo who have been stuck in their respective engines for like a decade now.

>>27860
I wouldn't count on it

__hatsune_….jpg - 226.29 KB (920x1200)

>>27755
Miku is my sister now

>>27796
To me this sounds like they use some sort of "AI resampler".

>>28497
i'm pretty sure hifisampler for utau already does something like this

This is probably the only way we'll ever get a new Kaai Yuki voicebank

>>28571
wouldn't that depend on yamaha not crypton?

>>28572
It depends on Yamaha, AHS and her voice provider. Word on the street says AHS actually wanted to make Yuki for SynthV but Yamaha didn't let them use the Vocaloid era recordings to train the voicebank.
So who knows.

>>30875
Half true. This comes from one of the AHS livestreams. The one for Miki and Kiyoteru’s announcement iirc. Their specific wording was that in order to create the bank they want, they would need to negotiate with Yamaha about using her Vocaloid 4 bank and data. There’s no confirmation that negotiations have happened. For all we know they got the okay, then the plan changed with SV2.

>>30894
It's gonna come out, trust