help me please 😿
#Help, I overtrained a model in RVC Disconnected
azure nova
Try out the generated models first, if it doesn't sound good to you, probably best to retrain with a cleaner dataset
pale escarp
The data set has no remains of instruments or noise, it is very clean, is there another way to correct it, it does not sound very good @azure nova
azure nova
What was the batch size you used? The lower the batch the more it should sound like the dataset
pale escarp
What was the batch size you used? The lower the batch the more it *should* sound...
the one that comes default 8 😿
pallid ploverBOT
Ayo? @pale escarp level 1 !!! 
azure nova
Try 4 or 6
pale escarp
I didn't know you could move that, thank you very much!
It's here right?
@azure nova
I have one last question, 4 would sound better than 6, right?
azure nova
Yup that's the one
It should be closer to what the dataset sounds like
Also meaning if you have stuff like laugh or cough it may add it to the model
pale escarp
I'll leave it at 4 let's see what, thanks @azure nova
azure nova
Aight goodluck
shut urchin
The data set has no remains of instruments or noise, it is very clean, is there ...
make sure you've cleaned/processed the dataset properly in order not to cause many mode collapses, as in the guide like this
https://discord.com/channels/1159260121998827560/1186762171204505801
including the audio labeling method based on silence threshold in Audacity that works like RVC's slicing in preprocessing stage but better
