How to use TTS voice to train a singing RVC model?

[whole bobcat]

I have some rough ideas and wanted some input from others. To train a talking RVC model I would need talking samples which generated tts voices can easily provide. However, to train a singing RVC model, I would need singing samples which generated tts voices cannot provide since tts cannot sing.

Would creating talking samples in various pitches and substituting them for singing samples be a viable method? Would this still allow the trained RVC model to sing properly?

[mighty prism]

you dont need singing samples for an rvc model to sing

[lyric knoll]

The model can sing, as long as the acapellas provided are close to the range of the model

You can do -12 in pitch whilst inferencing too

[mighty prism]

yeah fax

[whole bobcat]

as long as the acapellas provided are close to the range of the model
and to increase the range of the model, creating talking samples of various pitches would suffice?

[mighty prism]

i would say yes

[whole bobcat]

alright, thank you!

[mighty prism]

no worries

[lyric knoll]

Nah this wouldn't be that good

Just train with the samples you have and modify the pitch in the inferencing part

[mighty prism]

if he's manually adjusting the pitch

before training

that would damage the quality

if he just manage to found them

it should increase the pitch range

[lyric knoll]

Exactly my point

[mighty prism]

depends on the process

[lyric knoll]

That's why he should use the samples he can get, and change the pitch while inferencing an acapella

[mighty prism]

yeah fax

[whole bobcat]

I see what you guys are saying. That does make sense. I'll try it out first then. Thanks again!

[mighty prism]

good luck bro

[lyric knoll]

Alrighty, you're welcome, good luck on the model!

You can mark this as solved btw.