-
If you do not recieve your confirmation email within a few hours, please email haloutau@gmail.com with your username for manual validation. Your account should be activated within 24 hours.
You may also reach out via any other listed contact on Admin Halo's about page: https://utaforum.net/members/halo.194/#about
IS AI A GOOD WAY TO PRESERVE VOICES/UTAUs? (SURVEY)
- Thread starter Natsucchi
- Start date
Yes! As long as the creator of the original voicebank is consenting to it, it can work absolute wonders. Say there was an UTAU made in 2010 and the voicer of it came back ten years later- their voice certainly has changed, but they may still want to create voicebanks with their old voice. I can’t see why anyone would object to something as simple as that.
imho, no. I don't see the point in it, as if the VB is still available then there isn't a reason to """""preserve""""" it with AI.
Preservation comes more with ensuring the [thing] in question is made available in as many ways as possible, in it's original form. If it is edited, as it would be with AI, then it isn't being preserved due to the fact that it is no longer the [thing] in question.
Preservation comes more with ensuring the [thing] in question is made available in as many ways as possible, in it's original form. If it is edited, as it would be with AI, then it isn't being preserved due to the fact that it is no longer the [thing] in question.
I believe Hoshi meant training an RVC model to create new voicebanks for an older UTAU of whom the voice provider cannot voice again due to puberty/hormones/anything that would have a major impact on one’s voice.
And? My point stands. I, myself, have a myriad of UTAU I can no longer hit the pitch of from 10-15+ years ago, but I'd never consider using AI in order to "re-voice" them, nor create new VBs for them.
But, that's coming from someone who has negative interest in any AI whatsoever. So, whatever floats peoples boat, I guess.
DelphicVoiceAddragh
Teto's Territory
Honestly, I don't see a point in using AI to make new versions of old voicebanks. If you can't do the voice anymore, you can't do the voice anymore. No need to make a 'new' voicebank. It's not like there's been any updates to the engine that makes old voicebanks incompatible.
Besides, you're fundamentally changing the voicebank, not keeping it in original condition. It's not preservation, even under the loosest of definitions.
Besides, you're fundamentally changing the voicebank, not keeping it in original condition. It's not preservation, even under the loosest of definitions.
For preservation? No, not really. The only use for preservation I think makes sense is, for example, if you only have part of the data in the question, but most of it is lost media; and even then, it's not really ethical without permission from the original creator, and if the original data is lost, then it's likely the original creator can't be easily reached anyways.
It's not really "preservation" to say, use AI to make a brand new voicebank out of old sample data, when that old sample data is still available in its original form. If its original form is still accessible - then it's already preserved. I think it makes sense if ( again, keeping consent in mind ) you only have partial material available, and you want to use AI to restore it as best as it can, and try and make at least a rough approximation of the missing original material, just so there's something that resembles that missing material available. But it's not "preservation" to use AI to, say upgrade an older voicebank to a more modern standard. Generally speaking, that's just a whole new separate product as long as the original is still around.
I think it's fine if people want to do that with their voicebanks ( again, CONSENSUALLY, cannot possibly stress that enough ), but I think claiming it as "preservation" of vocal-synths comes across as a bit dishonest if it's not something like the very specific aforementioned scenario, and even in that case the ethics of it can be very dubious.
It's not really "preservation" to say, use AI to make a brand new voicebank out of old sample data, when that old sample data is still available in its original form. If its original form is still accessible - then it's already preserved. I think it makes sense if ( again, keeping consent in mind ) you only have partial material available, and you want to use AI to restore it as best as it can, and try and make at least a rough approximation of the missing original material, just so there's something that resembles that missing material available. But it's not "preservation" to use AI to, say upgrade an older voicebank to a more modern standard. Generally speaking, that's just a whole new separate product as long as the original is still around.
I think it's fine if people want to do that with their voicebanks ( again, CONSENSUALLY, cannot possibly stress that enough ), but I think claiming it as "preservation" of vocal-synths comes across as a bit dishonest if it's not something like the very specific aforementioned scenario, and even in that case the ethics of it can be very dubious.
Last edited:
1ceEatsStairs
Momo's Minion
in my opinion not really ?? i feel like the voicebank just existing itself is preservation as long as you hold onto it, AI sometimes really messes up with voices and can make stuff like accents get lost, plus I'm like really against AI so its an automatic no but not because "oh AI = bad" (AI can have a use, i just hate the people who use it for their own gain; (basically [quite literally] as a replacer, not a tool ) )
though it's not entierly preservation if you use AI to make a new voicebank to, since its not exactly the original.
I have zero issues if it's done with consent, but like AI voice stuff is really iffy honestly since you have the heavy pros and cons and how people use it.
though it's not entierly preservation if you use AI to make a new voicebank to, since its not exactly the original.
I have zero issues if it's done with consent, but like AI voice stuff is really iffy honestly since you have the heavy pros and cons and how people use it.
As someone who used RVC to AI for my new voicebanks for Alex, it really i guess just depends. I obviously could still do his voice, since I recorded the dataset in 2025, and I had lost my original voicebanks for him years ago due to the fact that im the King of lost media (i lose all my media)
Obviously making the voicebank without ones consent isn't cool, but if the parties are consenting then I dont see an issue. Considering a lot of companies for commercial vsynths are headed towards using AI datasets to train their voicebanks for clearer sounds, its kind of a norm now? It's the "ethical use of AI" in my opinion.
Though I will note that yes doing an AI voicebank for your UTAU does lose some of the pronunciation charm you'd have if you recorded the samples yourself. Though the quality is a lot smoother and the process a lot faster. So it's kind of like weighing out your pros and cons.
I don't have the same time I did, but I had always wanted to give Alex some new voicebank upgrades, so this was the best solution for myself. Because knowing me if I didn't, I'd be recording samples for months.
But again, and I can't stress this enough, the voicebanks need to be made with explicit consent from the voice provider. If a voicebank already exists and has easy access to getting, I don't see a point in making a new AI voicebank off their samples. In my case, though, this was a way of voicebank preservation because I had lost his voicebanks.
Obviously making the voicebank without ones consent isn't cool, but if the parties are consenting then I dont see an issue. Considering a lot of companies for commercial vsynths are headed towards using AI datasets to train their voicebanks for clearer sounds, its kind of a norm now? It's the "ethical use of AI" in my opinion.
Though I will note that yes doing an AI voicebank for your UTAU does lose some of the pronunciation charm you'd have if you recorded the samples yourself. Though the quality is a lot smoother and the process a lot faster. So it's kind of like weighing out your pros and cons.
I don't have the same time I did, but I had always wanted to give Alex some new voicebank upgrades, so this was the best solution for myself. Because knowing me if I didn't, I'd be recording samples for months.
But again, and I can't stress this enough, the voicebanks need to be made with explicit consent from the voice provider. If a voicebank already exists and has easy access to getting, I don't see a point in making a new AI voicebank off their samples. In my case, though, this was a way of voicebank preservation because I had lost his voicebanks.
heynotloid
Teto's Territory
Personally, I use UTAU as a way to preserve my voice. At any moment, I'm going to have throat surgery because some things that were operated on me when I was 3 years old have grown back, and when I do, I will have a much more mature voice, since at this moment, my voice has not matured almost at all since I was 8 years old. In that case, I consider that UTAU would be a good way to preserve my current voice for the day I have the surgery.
Couldn’t agree more.
You can actually preserve pronunciation details through RVC with index files, but it’s rather finicky.
I noticed when recording Alex's dataset in japanese vs English his pronunciation /was/ different. His soft voicebank had more of the accent I have when I sing things in japanese vs his standard vb which was recorded with an English dataset. That's the only reason why I said it can lose the accent. His pronunciation used to be horrendous, so im glad it cleaned it up a lot.
Similar threads
Similar threads
-
I wonder how to make voices more nasally in audacity- Started by Natsucchi
- Replies: 5
-
-
-
How do I use the "Pick up minimum voices for UST file" feature in SetParam?- Started by GeneralNuisance
- Replies: 1