When I listen to the sound it is often just not realistic. I've tried lots of models but none of them have worked.
#Need help making the rvc sound more realistic
lilac nexus
west ravine
When I listen to the sound it is often just not realistic. I've tried lots of mo...
elaborate:
- what u want to do
- ur pc gpu
- what tut link u followed
halcyon oasis
When I listen to the sound it is often just not realistic. I've tried lots of mo...
Wdym?
Tbh RVC doesn't sound very human
lilac nexus
Wdym?
im using w-okada and im trying to use it to pass as a human
im using the fork one uh
Deiteris fork
when it plays back the voice it feels like it shaky and it sometimes it makes a similar sound to the word but it isnt the word
lilac nexus
Tbh RVC doesn't sound very human
it doesnt? i thought it was good enough to pass as a human
silver jay
is your mic and background noise ok?
near oracle
Model dependent
lilac nexus
is your mic and background noise ok?
i dont have almost any background noise, maybe a small amount from fans
lilac nexus
is your mic and background noise ok?
is it usually good enough to pass as a human?
hoary topaz
is it usually good enough to pass as a human?
to pass as a human train a 2 hour dataset minimum
lilac nexus
can someone point me to a good model?
short fulcrum
to pass as a human train a 2 hour dataset minimum
10 mins of consistent dataset can also be decent, adding more data could be like jumping from 85% to 95% then 99% (in stability/generalization perhaps)
it may also depend on pretrain being used
hoary topaz
10 mins of consistent dataset can also be decent, adding more data could be like...
no, 10 mins will never sound human
its okay for training but for realistic results you need a bigger set
short fulcrum
no, 10 mins will never sound human
its okay for training but for realistic resul...
it can have less overall issues than longer inconsistent data
perhaps somewhat I stumbled on several song tracks that may have different loudness and EQ profile on vocals of the same singer
hoary topaz
it can have less overall issues than longer inconsistent data
perhaps somewhat I...
find a 5 hour dataset (twitch, yt)
train it using batch16
after that train only 10 minutes from the same set
you'll see
hoary topaz
so i need to find a very long trained model?
model trained with a big clean dataset and the appropriate batch size
hoary topaz
and whats a clean dataset?
no noise, weird sounds, etc
short fulcrum
ill try to find a 5-6 hour one
not always, but 30m ~ 1h shouldn't be bad
hoary topaz
agree
1 hour is already pretty realistic
not as realistic as a 2 hour or 5 hour model, but good enough
pulsar nebula
not as realistic as a 2 hour or 5 hour model, but good enough
Hello can you please tell me where can i find a 1hour+ voice models
short fulcrum
Hello can you please tell me where can i find a 1hour+ voice models
depend on your luck
pulsar nebula
depend on your luck
Really sorry im confused...but are these models like a rare epic drop loot? do i need like a good rng to get them?
short fulcrum
Really sorry im confused...but are these models like a rare epic drop loot? do i...
well, you seem to mean getting the existing models rather than the source to make dataset
unless you're interested on making datasets, please refrain from asking in someone else's thread
lilac nexus
unless you're interested on making datasets, please refrain from asking in someo...
nah its fine
ask away unwashedsocks
west ravine
@pulsar nebula it would be better to make your own help post, we have singular help post so we can track when a specific user issue is solved
@lilac nexus is your issue solved? do u need more help?
short fulcrum
nah its fine
if solved, let us lock this thread