search

RVC-Project / Retrieval-based-Voice-Conversion-WebUI

Easily train a good VC model with voice data <= 10 mins!
MIT License
19.7k stars 3.04k forks source link

Giving my voice model an accent #2029

Open EldritchSpiral opened 2 weeks ago

EldritchSpiral commented 2 weeks ago

So I'm working on a novel where one of the main characters is a singer and I decided to create a voice model for her so I can hear her sing in her actual voice. I actually have several "voice claims" I want to use, but I have a problem when it comes to her accent, since none of the voice claims have her nationality, and hence, not the accent I'm looking for. So essentially I'm curious is there any way to give my model a specific accent without changing her voice much?

78Alpha commented 2 weeks ago

If setting index ratio to 1 doesn't add enough accent, then you'd need someone already singing with the accent. RVC layers a voice on top, it didn't usually do such drastic alterations to the source.

EldritchSpiral commented 2 weeks ago

If setting index ratio to 1 doesn't add enough accent, then you'd need someone already singing with the accent. RVC layers a voice on top, it didn't usually do such drastic alterations to the source.

The thing is that for index ratio to work I need my model to have the accent already, which is exactly the problem since none of the voice claims I have for her have the accent I need. Would merging my result model with another voice with the accent be enough? What's the minimum weight a model should have in the merging process to pick up its accent while having mostly the timbre of the other model's voice?