Other Cryptonloid voicebank updates, collabs, & concert news (crypton_wat Twitter translations)
- Thread starter uncreepy
- Start date
I'm not sure, but, iirc I read somewhere about them wanting to make the newer voices sound closer to "how people originally interpreted them", so that sounds kinda like they'll be reusing samples and cleaning them up. I can't remember where I read this, though, so probably best to take what I say with a grain of salt. :S
As far as I know theyre re-using samples from V2 Miku, I think its safe to assume theyre reusing for the others too. Doesnt Crypton reuse samples all the time anyways? Even if they said they were fully re-recorded I would take it with a grain of salt from them
They didn't re-record at all. (I'm so sick of hearing people on Twitter saying Saki Fujita is still trapped in the recording dungeon when she hasn't recorded anything new for NT!)
They edited the original samples by making the pronunciation clearer/less muffled and shorter. The point is to preserve the way fans perceive their voices as supposed to be sounding because some think the newer versions don't sound close to them anymore.
Crypton edited the V2 voices for the Kagamines, Luka, and Miku.
Crypton edited the V3 voices for Kaito and Meiko.
The project was revealed by Wat showing a screen shot of Kaito's V1 voices getting edited, but at some point they switched to his V3 (I made a huge tinfoil hat timeline a while back and it says when the switch happened, but there's no way I'm re-reading this):
That's why I think Kaito should have been first to be released! Because he got mentioned first and I had my hopes up since before VocaloidOtaku died. orz Why couldn't he have at least been second... I think that the V3 Kaito and Meiko need the update the soonest, or Luka because of her EVEC not translating to Vocaloid Editor 5.
Edit: Also, I was convinced Luka would be first because of the way they kept saying the new voices were going to happen for Luka's anniversary. That was so embarrassing last year waiting for news all month with the 30 days worth of anniversary drawings and then nothing happened! I even made a Reddit post letting everyone know about it and then a year later, still nothing but a dumb beta Miku.
They edited the original samples by making the pronunciation clearer/less muffled and shorter. The point is to preserve the way fans perceive their voices as supposed to be sounding because some think the newer versions don't sound close to them anymore.
Crypton edited the V2 voices for the Kagamines, Luka, and Miku.
Crypton edited the V3 voices for Kaito and Meiko.
The project was revealed by Wat showing a screen shot of Kaito's V1 voices getting edited, but at some point they switched to his V3 (I made a huge tinfoil hat timeline a while back and it says when the switch happened, but there's no way I'm re-reading this):
That's why I think Kaito should have been first to be released! Because he got mentioned first and I had my hopes up since before VocaloidOtaku died. orz Why couldn't he have at least been second... I think that the V3 Kaito and Meiko need the update the soonest, or Luka because of her EVEC not translating to Vocaloid Editor 5.
Edit: Also, I was convinced Luka would be first because of the way they kept saying the new voices were going to happen for Luka's anniversary. That was so embarrassing last year waiting for news all month with the 30 days worth of anniversary drawings and then nothing happened! I even made a Reddit post letting everyone know about it and then a year later, still nothing but a dumb beta Miku.
I just want to remind everyone that Miku NT is supposed to come out this month during chuujun (11th ~ 20th range). I was looking at the Japanese Sonicwire campaigns and currently iZOTOPE is on sale until the 16th, so I wonder if Miku will be released around that time so people can still decide if they want to take advantage of the sale? When Miku NT was about to come out, they had a period where you could download Nectar 3 Elements from them for free Aug 2 ~ 12th (Jeez, that was page 9 of this thread!), that's why I feel like it's possible this new campaign is related to Miku NT's release.
Was just thinking about that this morning, actually. iZotope going on sale again seems very telling to me.
Hatsune Miku NT prototype is delayed, and now scheduled to release in early April
Not surprising, tbh, I kinda figured she'd be delayed. I hope everyone focuses on health and safety right now, and that Miku just gets done when she gets done.
Agreed, I kinda figured that might happen. CFM's probably short-staffed right now. And anyway, it might be in poor taste to try to get people excited about her when they're facing these other grave issues. Let's let Miku do what she does best and lift spirits and try to keep people safe and get excited about the new version when we can all be really happy about it.
(I don't mean to spam-reshare that post; I just think 1) it's important, 2) it's really cool--vocal synths really are "the people's idols," and 3) this is really what I'd hope Miku would be doing in a situation like this. That thinking is why the concerts/events got postponed, even before the governments said they had to be. )
(I don't mean to spam-reshare that post; I just think 1) it's important, 2) it's really cool--vocal synths really are "the people's idols," and 3) this is really what I'd hope Miku would be doing in a situation like this. That thinking is why the concerts/events got postponed, even before the governments said they had to be. )
I thought it would be canceled, Wat hasn't tweeted about the software since about the 12th of February.
Anyway, details from the blog post:
Miku NT is now scheduled for joujun (1st ~ 10th) for April.
What's included in Miku NT:
Miku's original voice bank
The following expression styles they are developing:
Auto Dynamics Control (beta), Attack Speed (for pronunciation control), Super Formant Shifter (beta) (effect), Wave Viewer, Direct Pitch Edit (new function).
I think that the Wave Viewer is the blocks of waves under the note boxes that says "Target".
Volume, Panpot
NT Parameters:
Consonant Rate
Attack Speed
Dynamics / Breathwave? (Is that what it says?)
Voice Voltage
Super Formant Shifter
Pitch Control
Voice Drive:
Rout Voice
Effect Level
The Sonicwire page updated, too:
Anyway, details from the blog post:
Miku NT is now scheduled for joujun (1st ~ 10th) for April.
What's included in Miku NT:
Miku's original voice bank
The following expression styles they are developing:
Auto Dynamics Control (beta), Attack Speed (for pronunciation control), Super Formant Shifter (beta) (effect), Wave Viewer, Direct Pitch Edit (new function).
I think that the Wave Viewer is the blocks of waves under the note boxes that says "Target".
Volume, Panpot
NT Parameters:
Consonant Rate
Attack Speed
Dynamics / Breathwave? (Is that what it says?)
Voice Voltage
Super Formant Shifter
Pitch Control
Voice Drive:
Rout Voice
Effect Level
The Sonicwire page updated, too:
Cool! So it looks like they're adding two additional effects relative to what they had before (there were previously 3, IIRC). So even though we're waiting longer, we're actually getting more stuff in the beta release. That seems like something to be happy about!
Piapro joins the likes of DV and CeVIO with direct pitch drawing!! If I wasn't already planning on throwing all of my money at Crypton, that alone would have convinced me, honestly.
I forgot Miku NT had a March release... But the April dates seem more finalized than the March one was. I'm curious to hear more demos prior to release but given everything right now, I'm satisfied waiting til end of the month or even release date for more demos.
Finally some news from Ryo.
Demo on iPhone of real-time voice conversion (male to female)
Left = Periodic Wave = voice input with a mic
Right = Aperiodic Wave = Voice converted to character's
The algorithm has a delay of 21 ms, so right/aperiodic wave/coverted voice is delayed.
Ryo thinks if he studies more, he will be able to increase the quality.
The voice being converted is from JVS Corpus (explanation link in English)
Reminder: This is Beef Jerky, it was initially Ryo's voice analysis software (similar to VocaListener to detect where to put notes but it didn't convert anything at the time. Now it appears he's turning the prototype into actual voice conversion software (like Cherry Pie/Voidol, but sounds a lot better). Based on the fact that all of the Beef Jerky demos have been vertical and he's revealing this is for iPhone, it must mean he intended it to be a phone app from the very beginning. Because Ryo is the head developer at Crypton for deep-learning/software development, I could see Crypton using this in the future to be like Voidol's phone version.
You can also see if you look at the following older posts about Beef Jerky that the interface has improved (loooks cleaner, has option for dry/wet sound):
Who is Ryo?
Meaning behind Cherry Pie's name
1st introduction to Beef Jerky
About Beef Jerky and the voice actor corpus
2nd Beef Jerky demo / comparison to VocaListener
Throwing in a post about Cherry Pie so you can see that the Beef Jerky interface looks like Cherry Pie's
Demo on iPhone of real-time voice conversion (male to female)
Left = Periodic Wave = voice input with a mic
Right = Aperiodic Wave = Voice converted to character's
The algorithm has a delay of 21 ms, so right/aperiodic wave/coverted voice is delayed.
Ryo thinks if he studies more, he will be able to increase the quality.
The voice being converted is from JVS Corpus (explanation link in English)
Reminder: This is Beef Jerky, it was initially Ryo's voice analysis software (similar to VocaListener to detect where to put notes but it didn't convert anything at the time. Now it appears he's turning the prototype into actual voice conversion software (like Cherry Pie/Voidol, but sounds a lot better). Based on the fact that all of the Beef Jerky demos have been vertical and he's revealing this is for iPhone, it must mean he intended it to be a phone app from the very beginning. Because Ryo is the head developer at Crypton for deep-learning/software development, I could see Crypton using this in the future to be like Voidol's phone version.
You can also see if you look at the following older posts about Beef Jerky that the interface has improved (loooks cleaner, has option for dry/wet sound):
Who is Ryo?
Meaning behind Cherry Pie's name
1st introduction to Beef Jerky
About Beef Jerky and the voice actor corpus
2nd Beef Jerky demo / comparison to VocaListener
Throwing in a post about Cherry Pie so you can see that the Beef Jerky interface looks like Cherry Pie's
Interesting! So there's likely going to be a mobile offering of some kind in the Piapro/CFM ecosystem, at least in the case of this particular voice conversion tool. Neat to see them branching out. It might be particularly useful for vtubers, etc., who could be getting material while out and about, doing a quick voice-converted commentary, and posting it.
Correct me if I'm wrong, but neither of the female voices in this video appear to be currently existing CFM Vocaloid characters. If that's the case, it seems that Beef Jerky isn't a plugin for the Vocaloid->Piapro gang but a separate project meant for a more general usage like what Voidol is supposed to be. I hope that's actually what's going on here because it's results sound way better than any of the competitors in that field, and I would love to see software like this available for people who want to disguise their real identities online like V-tubers.
I do hope there's a PC version though.
I do hope there's a PC version though.
The voices he used are available for developers to use from the Corpus website (contains like 5 hours of deep learning-related voice actress records or something close to that), so one of the characters from there is the voice you are hearing. Beefy Jerky is Ryo's pet project, but I don't think it would stay separate from Crypton very long, because Ryo is a very important worker there and he did most of the work on Cherry Pie (the interface and algorithms and most of the voice effectors were his idea/work). I think that what he is learning from this project is likely to be implemented in Cherry Pie in order to make it more accurate. Ryo doesn't tweet very often, but he sometimes reveals behind the scenes info relating to Cherry Pie and lately Beef Jerky.
