search

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

Easily train a good VC model with voice data <= 10 mins!
MIT License
23.06k stars 3.46k forks source link

Bad result train with a voice #916

Closed Fairplayn closed 4 months ago

Fairplayn commented 1 year ago

I've trained many popular voices perfectly, but for some reason and for this voice only it doesn't do a good job. https://whyp.it/tracks/113084/01?token=DVdVc (audio) I have 20 minutes dataset with this high quality audio.

ms903x1 commented 1 year ago

If your dataset consists entirely of this serious and formal broadcasting tone, then in order for the trained model to achieve good inference results, the input source should also mimic this broadcasting tone as closely as possible.

Fairplayn commented 1 year ago

well even if i use the same voice, it's not that good I want this voice: https://www.whyp.it/tracks/113084/01?token=DVdVc but the result is this(bad): https://www.whyp.it/tracks/113106/none?token=Lz00f

And for example i tried another voice like elon musk (works perfectly) elon musk voice: https://www.whyp.it/tracks/113107/elon?token=n6Q7I

Even If i change the tone it's the same problem, it doesn't look good for this voice only

github-actions[bot] commented 4 months ago

This issue was closed because it has been inactive for 15 days since being marked as stale.