DIY Voice Database Possible?

Synthesizer V Studio
1

Has there ever been any discussion of whether there might be a possibility of creating your own Do It Yourself (DIY) voice database?

I am a decent singer (within my genre) and I would love to be able to have the option of using my own voice in Synth V without actually singing every time! I could imagine this would be quite useful for prototyping songs as I write them. And maybe even to use my voice as a harmony part. And, as I get older and lose some of my singing capabilities I would still be able to “sing” my songs!

I’m sure this gets into Dreamtonics’ proprietary IP and could conceivably cut into their voice database sales so I understand this is probably not likely but I thought I would ask!

1 Like
2

Have you tried Vocoflex with your own voice as a sample input?
But I understand the need to use your own voice with SynthV.

1 Like
3

I’ve not seen that Vocoflex is something I need/want right now. And I wouldn’t want to convert my voice, rather, use it as a realistic version of me in Synth V.

4

This is a subject that has cropped several times in the past.
The methods used by Dreamtronics are closely guarded and unless you can afford the recording time and their fees will probably never be possible.
Several commercial projects have floundered, apparently because they were deemed not commercially viable.

You could try How do i become a voice provider? - #5 by Mechie

2 Likes
5

Thank you for that recommendation!

6

Creating a custom voicebank is expensive, so unless you’ve got a voice that fills a niche, I doubt that Dreamtonics would consider it. The actual cost is hidden behind an NDA, but the Kickstarter for Solaria was around $42,000.

I don’t have access to the inner workings of Dreamtonics, but I wouldn’t place any hopes on being able to create a custom voice in the immediate future. They’ve just released Vocoflex, and I suspect that wouldn’t have gone that route if custom voicebanks were easy to do.

If you really want to clone your voice, you’ve got at least three options, each with caveats:

  • SynthesizerV + Vocoflex
  • SynthesizerV + RVC
  • ACE Studio (obviously not SynthesizerV)

The biggest caveat is that much of makes a singer distinctive isn’t in their vocal timbre, but their performance. Just in the way that two people will read the same words differently, different singers will sing differently. If you’ve listened to bad RVC, you’ll know what I’m talking about.

And none of the options I listed captures the vocal performance.

So if you simply replaced a source singer with another singer’s timbre, it’s still going to sound very much like the original singer if they’ve got a distinctive style.

Vocoflex
Careful reading of the Vocoflex’s description will show it’s not voice cloning software. It’s intended to be used as a “creative” tool.

Still, it might work for you. Or not. The results for me aren’t great, because I’ve got a pretty low voice that doesn’t seem to match the Vocoflex vocal synthesis model very well. That’s fine, because - once again - Vocoflex isn’t really voice cloning software.

RVC
RVC is a popular method for voice cloning. It’s likely to sound more like you than Vocoflex - which doesn’t claim to do voice cloning - but how well this works again depend on how well the underlying performance matches how you sing.

The expensive bit of RVC is the training, which requires hours on GPUs, and laughably long times using a CPU. If you haven’t got a AI-capable video card, you can instead buy some credits to use Google Collab. Actually rendering can be done in realtime if you’ve got a reasonably fast CPU.

ACE Studio
ACE Studio does have an option for voice cloning. Like the other methods, it doesn’t capture performance. So again, the success of the cloning is going to be highly dependent on how much the source “sings” in your style.

Obviously, this is going outside the Dreamtonics ecosystem, but if an all-in-one voice cloning solution is what you’re after, they’ve got that.

Summary
With any option other than Vocoflex, you’ll need lots of voice data. ACE Studio recommends 30 to 100 minutes of audio for best results.

These are the options that I’m currently aware of that are available to you. For background vocals or replacing short segments of vocals, using RVC on a SynthV is probably the cheapest option.

Of course, all the is subject to rapid change as time passes. :laughing:

4 Likes
7

Wow, Thank You @dcuny for sharing all of that information! I really appreciate your generous sharing!!

8

As a first prototype for Solaria, I understand that it can be expensive. But after hundreds of voices with experience and tools, I doubt that for each voice they will invest the same budget. This is all the boss could imagine in order to be profitable on a long-term business. But of course, to keep the knowledge previously invested, they are not open to share it. But create a specific service for a highly motivated singer with reasonable financial means, why closing this option.

1 Like
9

I don’t believe that this is currently affordable to an individual with “reasonable financial means”. Even if it now took less than a quarter of the original cost, that’s still $10,000. And that doesn’t even account for the cost of the recordings, which is also substantial.

At the current cost, it doesn’t make sense for someone to create a SynthesizerV voicebank for themselves. After all, they’re a singer. So they don’t need a synthetic voice to sing for them.

For Dreamtonics, there’s an opportunity cost. Even if it doesn’t cost them $10,000 to do the work, time spent working on the voicebank is time they aren’t working on a different voicebank, or some other product. So it could very well be that the opportunity of working on something more profitable means it’s not worth their time.

If you’re just looking to clone a voice, RVC should be sufficient for your needs. I don’t know how ACE Studio does it, but if I were offering that as a service, that’s certainly what I’d use. :smile:

10

This is exactly the need raised by VocalDabbler. Singers (well, good ones) don’t need that… OK. Except when you’re old, sick, busy, etc.
The main users of DAWs are… musicians and not singers. So the need exists. And that’s exactly SynthV’s job. Capturing voices (your own, not the singer’s, etc.) could be quite interesting.
That’s why ACE, RVC already exist and also all the online AIs to use other voices and also yours. It’s not a dream, it’s a reality and a fact.
And vocoflex shows that it’s a need.
Maybe an advanced vocoflex option to integrate (like ACE) a complete voice and an export function to create a voice bank. With that you will use your voice directly without SynthV + vocoflex (with a result…well not complete) and with less CPU (thanks for the planet).

11

But the idea that you can create a voicebank cheaply and easily using Dreamtonic’s current technology is a dream, and not a fact.

You can argue otherwise, but that doesn’t make it so.

Vocoflex doesn’t claim to be voice cloning software.

You seem to be suggesting that there’s an easy way to create voicebanks for SynthV.

Creating a voicebank using for SynthV appear to require:

  • Hours of high-quality recorded vocals at multiple pitches with full phoneme coverage; and
  • Additional recordings of pseudo-singing to capture the singer’s pitch line attributes

At some point in the future, that might not be necessary. After all, it’s possible to train speaking voices on only a few seconds of recorded audio. But that requires a huge database of vocal data.

You can argue the technology should exist, and that Dreamtonics should be developing it. But the reality seems to be that it’s not as easy as you suggest.

1 Like
12

Yes for sure, that is not a simple job! But they have all the technology to do it. And others competitors (ACE…) will do it because of a… real need.
And I never suggested that something simple was an example.

Cheap not necessarily but much less than 40K$. Eclipse sounds and other bank designers would not launch into this market if they did not have a good idea of ​​the profitability of such a process. Between making perfect and making better, there is undoubtedly a market openned, ACE and AI is a perfect example of this market.

Oh yeah? The ability to import your own voice is one of its basic features. And all the registration precautions when purchasing this software are therefore unnecessary…

And I understand that a good voice bank is a huge process and not easy at the moment. But with time and current evolving technology (see AI developments), nothing is impossible. In the meantime, maybe a lower quality but usable voice bank could be interesting too.

1 Like
13

Well, you could read what Eclipsed Sounds has to say about that.

Actually, yeah. As in, that’s one of the things I talked about the only time I spoke with Kanru. That, and the usability issues with that color scheme. :smile:

I agree that the registration requirements don’t help lower expectations.

You can do that with Vocoflex or RVS right now.

14

Re-read the meaning of this first post… the needs are elsewhere.