I have a question about this...

[WoofMatcha]

I've too tired on re-recording many times of VCV voicebank because I want my voicebank to be compatible to most Japanese USTs that they're mostly using CVC, not CVVC and my CVVC voicebank sounds not so natural as my VCV voicebank while my VCV voicebank is always lacking of phonemes cuz I always pick efficiency-focused reclist.
So I have an idea to use AI RVC to do it for me, so my idea is to record my speaking tone. Then train it on Google Collab (cuz my laptop is weak for this AI stuffs), then give it the sample recording of other voicebank to make them sounds like me.

Is it safe and ethical? I mean on the part to use other voicebank's samples to change the voice into my voice? For my own voice I'm consenting on making my own voice into RVC AI, but I definitely will never publish it cuz I don't want people misusing my voice.

I just conflicted about it cuz I've been so tired on re-recording a lot of samples that just mispronounced and also some are messed up...
I'll reuse my voicebank's recordings as the training data for my own RVC model.

Or like this: use my own AI voice for testing a reclist, then once the result is decent.. I'll re-record it with my own real voice? Cuz I know some peoples might get upset with AI stuffs, so I'll try my best to not release the voicebank that using my AI RVC voice

 

[Kiyoteru]

Check the terms of use for the specific base voicebank that you're converting the samples from. If you can't find the answer there, contact the creator of the base voicebank.

For example, Namine Ritsu has a very permissive terms of use, so you would be free to use those UTAU samples as base audio to convert with your voice model.

 

[WoofMatcha]

Check the terms of use for the specific base voicebank that you're converting the samples from. If you can't find the answer there, contact the creator of the base voicebank.

For example, Namine Ritsu has a very permissive terms of use, so you would be free to use those UTAU samples as base audio to convert with your voice model.

Ok, I might check on it. Because I'm just doing it for testing, I definitely will not use it for long time too cuz I still want to make my voicebank made with my own voice, human voice. Not AI.

 

[Kiyoteru]

Voice conversion models can still be very useful for UTAU development. For example, you might create a model based on samples of your own voice from when you were younger, or models based on voice acting your own characters. Later, you can record your own base samples according to an UTAU reclist, then convert those samples to sound like the old voices.

AI is not inherently evil. The main ethical issues are consent for the data used and power consumption. By using your own voice for both the training data and the samples to be converted, you have the full rights to use and modify the audio as you wish. If you can invest in your own hardware for doing all of the processing, such as building a PC and installing solar panels on your house to power it, you can do these intensive computations in an environmentally friendly way.

 

[SunnyWolves]

CVC? How are you going to do a CVC Japanese voicebank? I've only heard of English CVC. Do you mean VCV?

 

[WoofMatcha]

CVC? How are you going to do a CVC Japanese voicebank? I've only heard of English CVC. Do you mean VCV?

Ah yeah, typo..

Post automatically merged: May 2, 2025

I meant VCV.

 

[SaKe]

RVC is just a fancy voice changer. It’s entirely ethical to use as long as you aren’t trying to impersonate someone else.

As for voicebanks you can RVC, Namine Ritsu’s TOS is essentially ‘do whatever you want bro’ and PIERO allows RVC conversion too. For the latter, though, his VCV voicebanks are all unreleased but I can send them to you if you’d like.

 

[WoofMatcha]

RVC is just a fancy voice changer. It’s entirely ethical to use as long as you aren’t trying to impersonate someone else.

As for voicebanks you can RVC, Namine Ritsu’s TOS is essentially ‘do whatever you want bro’ and PIERO allows RVC conversion too. For the latter, though, his VCV voicebanks are all unreleased but I can send them to you if you’d like.

My plan is actually just for testing reclists, cuz there's so many reclists and I've tired on re-recording all of them. And most of them are just missing some phonemes that was used on the songs that I want to cover.

Even the simplest song as Ochame Kinou and "Song of Eared Robot" LMAO

I've checked the oto though, it's missing the specific phoneme LOL

The test using other voicebank's as "guide audio" is also part of my experiment too.

Post automatically merged: May 2, 2025

I maybe might also try using trial Synth-V to make "guide audio" to do the reclist pronunciations.

 

[SaKe]

You could also use COEIROINK or Voicevox, which would be easier than SynthV Lite. COEIROINK is a little higher quality than Voicevox but Voicevox has easier parameter control.

 

[WoofMatcha]

Finally the training is done in Google Colab, the model is done but...
Here's a catch: my laptop is not strong for this thing cuz no NVIDIA GPU (cuz they're expensive LOL)

Bruh, it's a problem for me...
Maybe I'll wait until the day I'm on my mood to record again when I'm recovered...
I really need to find more reclists that are complete cuz I want my voicebank to have complete phonemes (or maybe if there's a way to make a copy of existing ones with the samples that are sounds similar?)

 

[Kiyoteru]

I'm not sure which VCV reclists you've tried using before, but this website has a complete collection of japanese and english reclists.

JP Reclists | Wasteland UTAU

wastelandutau.neocities.org