VocaloidOtaku.net Forums - Providing Everything Vocaloid: Voctro Lab's New Singing Synthesizer - VocaloidOtaku.net Forums - Providing Everything Vocaloid

Jump to content


Welcome to VocaloidOtaku!

You are currently viewing our forum as a guest which means you are limited to some discussions and certain features.
Take a few minutes to browse around. Should you enjoy what you see, register and you will gain access to more stuff.

Registration is simple and fast. It won't fetch you more than a minute.
Click here to join!
Guest Message © 2017 DevFuse
  • (2 Pages)
  • +
  • 1
  • 2
  • You cannot start a new topic
  • You cannot reply to this topic

Voctro Lab's New Singing Synthesizer (Not yet available commercially)

#1 User is offline   D!zzy Icon

  • [dZ]
  • Icon
  • Group: Moderators
  • Posts: 3,753
  • Joined: 10-June 12
  • Gender:Female
  • Location:Spaced out
  • Producers:Hachi, Team Kamiuta, Ryuryu

Posted 14 April 2017 - 12:17 PM

As some of you are well aware, Voctro Labs has been continuing their research into singing synthesis. There most recent outing is a neural parametric synthesiser.

From the technologically minded, here's an explanation:
Spoiler



Features: English Male, English Female, Spanish Female Soft (MAIKA) and Spanish Female Power (MAIKA)

It sounds like Voctro Labs continues to be on the cutting edge of synthesis technology. I hope that these samples give an indication of what singing synthesisers may sound like in the years to come.
Posted Image Posted Image Posted Image

三月は夜の底 古川本舗

#2 User is offline   D Smolken Icon

  • Papa Ork
  • Icon
  • Group: Members
  • Posts: 220
  • Joined: 27-March 15
  • Gender:Male
  • Location:Poland

Posted 14 April 2017 - 01:20 PM

Yeah, there's a lot of progress on the research front. The samples sound quite natural and intelligible, too. A usable instrument using this kind of technology might come fairly quickly, too... one big change in the wider world of software development is shorter cycles and getting products to market ASAP.

#3 User is offline   missy20201 Icon

  • was Lεօղ
  • Icon
  • Group: Members
  • Posts: 943
  • Joined: 16-August 12
  • Gender:Female
  • Location:USA

Posted 14 April 2017 - 03:12 PM

These sound really good! I wonder if they'll end up trying to sell it? Certainly, if they ever did and included Maika (especially with multiple banks like with her regular and soft, like with these samples), I'd certainly be interested in picking it up when I'm not broke
The squad:
Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Posted Image Cyber Songman
Still need to be hazed in:
Posted Image Posted Image Posted Image
Sprites by haneoka c:

Owner of
Posted Image

#4 User is offline   FlyingCarParts Icon

  • VSTi Metal Musician
  • Icon
  • Group: VO+ Members
  • Posts: 216
  • Joined: 27-April 16
  • Gender:Male
  • Location:San Antonio, Texas, USA
  • Producers:owlscillate, cepheid, furez

Posted 14 April 2017 - 03:55 PM

Now I hope they get into metal vocals!

Ha... yeah right like that'll ever happen.
My Soundcloud Profile | YouTube | Facebook (faster PM replies) | Doodles

Windows 8 | FL Studio 11 | Piapro Studio | Paint Tool SAI

I designed Marie Ork's default growl art.

Vocaloids I own:
Spoiler


Alter/Ego voicebanks I own:
Spoiler

#5 User is offline   lilravn Icon

  • Cyber stalks better than most....
  • Icon
  • Group: Members
  • Posts: 1,736
  • Joined: 01-April 14
  • Gender:Female
  • Location:In denial...

Posted 14 April 2017 - 03:59 PM

Woah... The 2 Bottom tracks.. In spanish.. Are real amazing... The english ones suck...
Posted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted Image

My latest attempts at hyper-realism....

My 5 BEST EVA renders.......
Spoiler

#6 User is offline   D Smolken Icon

  • Papa Ork
  • Icon
  • Group: Members
  • Posts: 220
  • Joined: 27-March 15
  • Gender:Male
  • Location:Poland

Posted 14 April 2017 - 05:01 PM

View PostFlyingCarParts, on 14 April 2017 - 04:55 PM, said:

Now I hope they get into metal vocals!

Ha... yeah right like that'll ever happen.

We've proven that it's possible using technology designed for singing. This particular technology "significantly reduce(s) training and generation times". That could also significantly reduce the development costs of a new voice, making vocals that have a smaller niche commercially viable.

So, it could well happen, similar to how Toontrack or Drumdrops make very genre-specific drum kits. I'm sure they're getting lots of "can you do this with my voice" requests, but...

#7 User is offline   TJ Studio Icon

  • Vocal Synth Electronic/EDM Producer
  • Icon
  • Group: Members
  • Posts: 206
  • Joined: 15-September 16
  • Gender:Male
  • Location:Ringgold GA
  • Producers:VocaCircus

Posted 14 April 2017 - 05:07 PM

So. The MAIKA developers are making their own singing synthesis program

Maybe I'll try that, just gotta figure out how much that program costs
[Sprites By: BambooGarden101 - With one sprite by: Kidgore]


My Vocal Synth Voicebanks [In order of purchases]
Spoiler


Soon to be in the team [Not in order]
Spoiler


My Social Medias


Fanclubs I'm In
Spoiler

#8 User is offline   D!zzy Icon

  • [dZ]
  • Icon
  • Group: Moderators
  • Posts: 3,753
  • Joined: 10-June 12
  • Gender:Female
  • Location:Spaced out
  • Producers:Hachi, Team Kamiuta, Ryuryu

Posted 14 April 2017 - 05:59 PM

View PostTJ Studio, on 14 April 2017 - 05:07 PM, said:

So. The MAIKA developers are making their own singing synthesis program

Maybe I'll try that, just gotta figure out how much that program costs

Haha, this isn't their first time either. In fact, Voctro Labs was founded by the same team that developed the original VOCALOID engine!
Posted Image Posted Image Posted Image

三月は夜の底 古川本舗

#9 User is offline   Robyn Is A Ninja Icon

  • Wind Turbine of the Jellyfish Mom
  • Icon
  • Group: Members
  • Posts: 756
  • Joined: 09-February 14
  • Gender:Female
  • Location:Elsewhere looking for oarfish
  • Producers:AdyS, Yukitsuki, EmpathP, KTKT

Posted 14 April 2017 - 09:35 PM

Really impressive! I'm curious about what they're going to do with the technology and if it'll be available to the public. It'd be nice to have some realistic alternatives to Vocaloid. I was really impressed with the English samples, though the Maika samples are incredible (that Power bank, yo).

I wonder if this has anything to do with that mysterious voice sample that popped up a couple years ago? The one with the male Jazz voice.

The Squad in Order of Purchase:
Posted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted Image

Soon to Join:
Posted ImagePosted ImagePosted ImagePosted ImagePosted ImagePosted Image

ADORABLE sprites by the wonderful BambooGarden101! :D

Posted Image
Posted Image
Posted Image
Posted Image
Posted Image
Posted Image
Posted Image
I also own the WIL fanclub but I don't have time to slug through the slow servers of VO to get a new banner for my WIL Club

#10 User is offline   Zoku Icon

  • im a waking mess
  • Icon
  • Group: Members
  • Posts: 290
  • Joined: 19-February 12
  • Gender:Male
  • Producers:DystoP, kuben, CircusP

Posted 15 April 2017 - 12:30 AM

While I hope they commercialize this somehow, I doubt they will. I vaguely remember one of the lead Voctro Labs people saying that commercializing the Vocaloid engine was a mistake. I could be wrong, though.
deviantART | Tumblr | Soundcloud

I like UTAU! I'm also an amateur producer c:

#11 User is offline   _caustic_ Icon

  • Icon
  • Group: Members
  • Posts: 16
  • Joined: 09-August 16
  • Gender:Male
  • Producers:Kenji-B, Circus, DystoP

Posted 15 April 2017 - 03:26 AM

Has anyone linked this yet?
http://www.dtic.upf....M_NIPS_seminar/

It's the same thing just more information more or less if I'm not mistaken.

especially the poster linked on the same page.
http://www.dtic.upf....iles/poster.pdf

#12 User is offline   D Smolken Icon

  • Papa Ork
  • Icon
  • Group: Members
  • Posts: 220
  • Joined: 27-March 15
  • Gender:Male
  • Location:Poland

Posted 15 April 2017 - 06:19 AM

Specifics... the source dataset is just 10 minutes of speech. That's, what, 50 MB of mono 16-bit audio. Less than 10% the size of a V4 Vocaloid download. To get that level of quality with so little data really is a huge step forward.

I really like stuff like this, or the SampleModeling virtual instruments, which take a much smaller amount of data than typical sample-based instruments, and then milk as much as possible out of it. Not that this is any easier or anybody can do it - SampleModeling was founded by a cardiologist, who's done plenty of looking at repeating heart waveforms...

#13 User is offline   _caustic_ Icon

  • Icon
  • Group: Members
  • Posts: 16
  • Joined: 09-August 16
  • Gender:Male
  • Producers:Kenji-B, Circus, DystoP

Posted 15 April 2017 - 08:54 AM

And since the final result is a model it could be (would be) even smaller than that.

My only concern is that it's still not as "crisp" as the sample based voices. It certainly is more consistent than the HMM voices however.

Instrument modeling technically already exists but doesn't seem to be widely used, maybe it's just not seen as necessary? Though I can't imagine something designed for speech would readily accept instruments.

It states in the poster (not explicitly/specifically but it mentions it) that it was rendered using Nvidia gpus so it's possibly quite computationally expensive.

#14 User is offline   D Smolken Icon

  • Papa Ork
  • Icon
  • Group: Members
  • Posts: 220
  • Joined: 27-March 15
  • Gender:Male
  • Location:Poland

Posted 15 April 2017 - 10:45 AM

Instrument modeling is a whole another subject, yeah, but it usually means smaller size and higher CPU use, with stuff like Pianoteq or MODO Bass which are (almost) totally synthesized based on models.

Very low CPU use for rendering would be great (even if creating the model takes a lot of CPU resources) - if a normal computer could synthesize two dozen voices at once, we could do a live synthesis of a choir. THAT would be massive, because real choirs are expensive and logistically complicated to record, and sample-based choirs that can sing any lyrics are awkward to use.

#15 User is offline   FlyingCarParts Icon

  • VSTi Metal Musician
  • Icon
  • Group: VO+ Members
  • Posts: 216
  • Joined: 27-April 16
  • Gender:Male
  • Location:San Antonio, Texas, USA
  • Producers:owlscillate, cepheid, furez

Posted 17 April 2017 - 06:55 AM

View PostD Smolken, on 14 April 2017 - 11:01 AM, said:

We've proven that it's possible using technology designed for singing. This particular technology "significantly reduce(s) training and generation times". That could also significantly reduce the development costs of a new voice, making vocals that have a smaller niche commercially viable.So, it could well happen, similar to how Toontrack or Drumdrops make very genre-specific drum kits. I'm sure they're getting lots of "can you do this with my voice" requests, but...


True, but.

Look at how long it took for us to get a metal vocal library that growls in English.
As far as I know, the number of metal singsynth voicebanks out there of any language could probably be counted on one hand.

It's a very difficult thing to anticipate when:
-the people working on this tech don't seem to care to cater to metal vocals
-metal vocal libraries don't even seem to be commercially viable looking at stats since not enough metalheads are also singsynth fans (and there arent a lot of Marie/Maiko users, are there?)
-metal vocals might even require huge additional engine changes that the team doesn't see as necessary (although this is from my standpoint since I don't have deep knowledge of how this stuff works)
-there are actually a lot of types of metal vocals, not just growls (false chord, fry, inhale, squeals, etc.)

Metal with singsynth vocals is a very problematically niche interest...
My Soundcloud Profile | YouTube | Facebook (faster PM replies) | Doodles

Windows 8 | FL Studio 11 | Piapro Studio | Paint Tool SAI

I designed Marie Ork's default growl art.

Vocaloids I own:
Spoiler


Alter/Ego voicebanks I own:
Spoiler

  • (2 Pages)
  • +
  • 1
  • 2
  • You cannot start a new topic
  • You cannot reply to this topic


2 User(s) are reading this topic
0 members, 2 guests, 0 anonymous users