Synthesizer V - Vocaloid haters might want to check this

[AnhrithmonGelasma]

Vocoflex | Dreamtonics株式会社

dreamtonics.com

Vocoflex is Dreamtonics’ experimental
approach to voice morphing.

a tool to replace your vocals with any voice.

Vocoflex is a real-time voice morphing plugin which can transform a vocal recording to sound like one or more other voices, including the ability to combine voices.

To avoid unethical uses of the technology, the output will include an inaudible watermark which can be used to trace a sample back to the user who created it. This mitigates the risk of copyright violations, nonconsensual use of someone’s voice or property, and impersonation.

Have they indicated whether they'll permit use of this with Synthesizer V voice banks?

 

[raidergale]

I'm a bit iffy on the ethical implications of this, but it's good they're watermarking the results to quickly figure out who made it if someone misuses the program.

Have they indicated whether they'll permit use of this with Synthesizer V voice banks?

You can freely use Dreamtonics' SynthV libraries as input to morph from, other companies' voicebanks are a no go unless they specifically give the go ahead in their TOS.
If you're talking about using SynthV voices to morph to, then that's a no, since the TOS states that you can't try to replicate the tone of an existing voicebank

 

[AnhrithmonGelasma]

If you're talking about using SynthV voices to morph to, then that's a no, since the TOS states that you can't try to replicate the tone of an existing voicebank

But I imagine they might make an exception for use with their plugin, provided you've bought the voicebank (and they can get the voice providers etc. to agree, if Dreamtonics hasn't already been granted the rights to do that via whatever contracts were signed previously). Having a built-in library of quality voices would also make the plugin more appealing.

 

[AnhrithmonGelasma]

One issue with using Synthesizer V's current voice conversion is that it doesn't reflect timbre (or changes in timbre) or iirc dynamics, whereas RVC does. This new "morphing" plugin is really a voice changer.

And the TOS last I checked only prohibit using the Synthesizer V voices to train AI models. They don't prohibit, for example, morphing a vocal sample into an audio sample of a Synthesizer V voicebank using Morph2 or other software that doesn't train AI to duplicate the voice. The obvious intent is to prevent potential Dreamtonics competitors from training on their voicebanks.

 

[David Cuny]

I just signed up (but wasn't necessarily accepted) to the Vocoflex beta. Here's the EULA, in case anyone is interested:

Vocoflex EULA | Dreamtonics株式会社

dreamtonics.com

As others have said, this is a realtime vocal changer. It's likely that Dreamtonics trained it on a large variety of voices, so when a new voice is loaded, it determines where those vocal attributes lie relative to the learned voices.

I know a number of people who would be interested in this - they are using SynthesizerV because their own voice is no longer up to the task, but they've got plenty of raw vocal stems of themselves which could be used to create vocal clones.

I'd prefer that something like this would ultimately be built into SynthesizerV, since that's where I do most of my vocal editing. Hopefully they'll have that as an option, assuming it works well.

 

[David Cuny]

Having a built-in library of quality voices would also make the plugin more appealing.

But what does "quality" mean?

Many of the features we associate with a professional singer have nothing to do with the timbre of the vocal. For example, smoothly moving from note to note, staying on pitch center, appropriate vibrato and breath support are just some of the attributes of a good singer.

But all of these are "baked into" the base vocal, and RVC doesn't change any of them. That's what pitch correction software does. But even that's limited, because it's not entirely trivial to remove the singing errors with pitch correction software. If a singer's got a bad vibrato, you can flatten the pitch deviation, but other timbre changes won't be removed.

The plugin does have a feature where you can adjust parameters to create a "new" voice from scratch. I suspect that's what the plugin is doing with the source audio: determining which parameters get set where, relative to some "generic" voice. It then does the same thing with the source singing voice, so it can match the parameters to the target.

I'm not disagreeing - I'm just not sure how much can be expected of the plugin.

One use case comes to mind: fixing vocals that can't be corrected by pitch correction. Take vocals that need correction, and import them into SynthesizerV, edit them using a comparable vocalist, export to a .wav file, and use Vocoflex to change the voice. In theory, you should be able to seamlessly replace the vocal flubs.

Heck, you could potentially use that to correct phonetic issues that voices like Sheena have.

 

[AnhrithmonGelasma]

But what does "quality" mean?

Many of the features we associate with a professional singer have nothing to do with the timbre of the vocal. For example, smoothly moving from note to note, staying on pitch center, appropriate vibrato and breath support are just some of the attributes of a good singer.

In contemporary popular music (and related genres) pitch correction and heavy compression are widely used, and even expected, so this isn't so much of an issue (for those styles... personally I strongly prefer a lot less pitch correction and a lot more dynamic range than is typical these days). Of course it would be more ideal to have AI that could improve the singing in various ways (perhaps as an n-dimensional space you can navigate through with XYZD...etc sliders).

 

[Anthony]

If Vocoflex works very well, it could hurt their voice database sales. Users could, in principle, buy just a few 'cardinal' voices occupying the outskirts of singer voice-space axes and then morph between them to create markedly new voices (without having to buy additional voice databases).

 

[Stav7]

Not much tweaking... this is the total thing (so far) -



Voice is in auto mode, so other than the vibrato, there's no pitch alteration. Notes are placed on a 1/16th quantized grid. No customization of phonemes.

I have Izotope Nectar as an insert... its not really doing much, just a little compression and minor EQ - could use any channel strip really. I have two reverbs on there (an EMT plate, and an instance of Raum). Its mostly just leftover from noodling. Not much thought put into it at this point.

I was really just noodling around, and it kind of fell into this. I just had to post because the voice is so gorgeous and I feel as though Sheena has been overshadowed. I may have contributed to this overshadowing, since I wasn't really very enthusiastic about her when I got the voicebank. These days she is without hesitation, my favourite.

can you do the same sample without the reverb on the voice so I get an idea what the voice is like flat

 

[richiebee]

can you do the same sample without the reverb on the voice so I get an idea what the voice is like flat

This is it with absolutely nothing on it... no nectar, no reverb.

View attachment that-voice-bare.mp3

 

[Stav7]

This is it with absolutely nothing on it... no nectar, no reverb.

View attachment that-voice-bare.mp3

thanks, yes its a very nice voice, little more breathier than Solaria, which would definitely suit some of my songs

 

[richiebee]

thanks, yes its a very nice voice, little more breathier than Solaria, which would definitely suit some of my songs

Remember that at the start of this, the Soft vocal mode is pretty high. That would contribute to the breathier sound. Even Solaria is capable of pretty breathy sounding vocals if you push the Soft vocal mode.

 

[chapbot]

 

[bosone]

things are proceeding very fast!

 

[David Cuny]

Very cool.

There's an additional cost to create a custom voice in Ace Studio. I can't find the price listed on their website, as I don't have a subscription so I can't go past the initial screens.

They describe these voices as useful for "vocals that are necessary but not that crucial", so it makes me suspect the quality of the vocals. However, there don't seem to be any obvious issues with the example in the video.

Since Ace Studio is a flat-rate subscription model, adding support for custom voices doesn't cause issues with their revenue stream. In fact, it enhances it since it gives people more incentive to continue using their product.

On the other hand, SynthesizerV likely makes the bulk of their money selling voices. Allowing users to freely create custom voices would threaten that revenue stream.

I suspect a workable model for SynthesizerV would be to charge a flat rate for creating the voice, like Ace Studio does, and then locking the use of that voice to a single license.

 

[chapbot]

Very cool.

There's an additional cost to create a custom voice in Ace Studio. I can't find the price listed on their website, as I don't have a subscription so I can't go past the initial screens.

They describe these voices as useful for "vocals that are necessary but not that crucial", so it makes me suspect the quality of the vocals. However, there don't seem to be any obvious issues with the example in the video.

Since Ace Studio is a flat-rate subscription model, adding support for custom voices doesn't cause issues with their revenue stream. In fact, it enhances it since it gives people more incentive to continue using their product.

On the other hand, SynthesizerV likely makes the bulk of their money selling voices. Allowing users to freely create custom voices would threaten that revenue stream.

I suspect a workable model for SynthesizerV would be to charge a flat rate for creating the voice, like Ace Studio does, and then locking the use of that voice to a single license.

It's $300 for ONE custom voice!!

 

[mothershout]

Since I just bought Sheena and immediately tested her out on the vocal line for a new project... yeah, I do hear at least one Japanese-accent issue. Specifically, the default r for the Soft voice mode has a definite Japanese quality. But with Sheena I found that changing to an alternate for that phoneme made an audible difference; the Alt 1 r sound solved the problem. I usually find that phoneme alternates have no audible effect; I hope this is a sign of Dreamtonics offering more variation.

I'm going to follow up on my own post...

Sheena's a good voice for a backup, but like other Dreamtonics voices, her accent and other phoneme issues look (to me) like they rule her out as a lead. For example, I cannot get her to sing a word with a trailing "s" (such as remains) in any voice mode. The accented r isn't always fixable (depends where in a word it occurs) and I'm also hearing issues with trailing t or d sounds.

So yeah, I think I'm going to stop buying Dreamtonics' voices until they release one based on a singer whose native language is English.

 

[mothershout]

Remember that at the start of this, the Soft vocal mode is pretty high. That would contribute to the breathier sound. Even Solaria is capable of pretty breathy sounding vocals if you push the Soft vocal mode.

And, of course, you can always up the Breathiness parameter value as needed.

 

[elucid]

Remember that at the start of this, the Soft vocal mode is pretty high. That would contribute to the breathier sound. Even Solaria is capable of pretty breathy sounding vocals if you push the Soft vocal mode.

Can't thank you enough, @richiebee, for the quality demos of Sheena! She has proved perfect for a song I'm working on where Solaria wasn't quite getting there for me.

 

[Polarity]

I got Synthesizer V PRO and Solaria Full a week ago and I'm very satisfied with the experiments till now with the Extract Notes from Audio function in the ARA plugin; I'm using Cubase 11 PRO.
I'm still in the phase of studying manual and tutorials (on youtube) of the functions and tricks of software, but I already tried it over some old vocals for a cover/remix (Mysterious Times by Sash, it is a famous Trance-Dance track released in 1998) made by my friend in 2005, that wasn't so much a good performance that time, but it's not an easy song to emulate the original singer: anyway even if not finished yet to working on, I can say that the result is really amazing!

Of course I had to correct some words and note length, but it retained very well the original performance I gave it to convert: really... it works great!
From many demo-songs I heard online, I was sure that Solaria was a great choice as voice for the kind of music I do.
It worked perfectly also converting spoken words made as an intro: it simply seems my vocalist with a slightly different timbre and accent (sometimes), but her tone and emphasis are there.

Yesterday I also tried over an old vocal by me (made for make the vocalist learn the parts) of a song made with my group of that time and I was really impressed with the Solaria performance... while giving a listen to all my other takes I didn't know if laugh or cry.
Of course even with that I fixed words, length and pitches because I'm tone deaf as a bell, like we say here in Italy.

For now I'm very happy to have preferred Synth V to ACE Studio (I hate subscriptions) and now that I finally migrated to Windows 10 and the Extract from Audio was implemented the right time was
arrived!