#Is RVC the best model type for voice models?
1 messages · Page 1 of 1 (latest)
well for speech most tts are already far superior to rvc in terms of quality and realism
for realtime inference? there's only rvc and ddsp-svc, but the latter is harder to use and the developer only provides help if you speak chinese
(rvc is actually abandoned since 2023, but it has an active community up to this day, the original devs of the project moved on)
Wait can you use RVC models for tts? And if so is that why sometimes in the samples at quailty seems good but when j put it into a voice changer the quality significantly drops?
nono, tts models are superior to rvc thats what i meant to say
rvc has downgraded quality intentionally by the original main dev of the project, he did this so people can train models with weak gpus
ah and about the model sounding slightly different in realtime, thats normal, the realtime inference is weird
but it shouldn't be a massive difference
if ur getting such huge difference there's three options:
- model sucks
- your volume settings are bad
- low extra chunk value (below 2.7)
99% of the time is option 1
rvc was never intenteded to be used as a voice changer, rvc does NOT mean realtime voice changer
originally it was made for funny ai covers
the realtime thing is a hack someone else made (basically converting the local inference to realtime)
Oh okay
so its just up to me to find a good model then =w=
questions regarding mental sanity will not be answered today
explain what kind of issue that thousand models you've tried suck for you
maybe with some sample output audio
omg another saba fan!! But anyways its not that they suck a lot of them are actually pretty good theyre just not good enough to be convincing
again not good in what?
can't handle laughing, for example?
maybe you should lower your expectation and acknowledge the RVC limitations