Model sounds nothing at all like dataset

[charred snow]

I made a model with Applio but when I infer any audio, it sounds nothing like the dataset. Could just be the dataset but I'm not quite sure since I tried retraining 3 times with the same settings and they all sounded different

[gray chasm]

Might be your dataset yeah, what's its length and what batch are ya using

[charred snow]

It's about 6 1/2 minutes, and i'm using a batch size of 12

wait let me check

7 minutes actually

[gray chasm]

Try using batch 4, see what happens

[charred snow]

Alright hold on

[simple sable]

Can you send a sample of your dataset and the result audio?

[charred snow]

so like just 1 clip from my dataset and 1 clip from an output audio made by the model?

[simple sable]

Yup

Make sure so music btw

[charred snow]

ok

[rotund aurora]

Ayo? @charred snow level 4 !!!

[charred snow]

nvm it fixed it thx

[simple sable]

ok wtf is that difference i was expecting something else

[earnest sand]

dont use rmvpe_gpu

[gray chasm]

wha

[stable lion]

what?

why not

[earnest sand]

?

[stable lion]

rmvpe_gpu is the same as rmvpe it just uses gpu

why should you not use?

[earnest sand]

According to my experience, in some cases it affects the DATASET

[silver plume]

It shouldn't