This audio sample was inferred on the Aesthetic RVC Inferrence Hugging Face Space by r3gm using the RMVPE+ pitch extraction algorithm with "Split Audio" enabled on a minimum value of 55.2 Hz and a maximum value of 140.1 Hz. The audio sample also has the vocal track formanted male to female and it's pitch raised by 12.

87k steps???

Yes.

what gpu did you use

imma guess you have rtx 40-series

I was training on a Colab; I didn't have a GPU installed on my computer. I used the Tesla T4 with a batch size of 8 when training.

oh

you have colab pro or something?

because damn

that took like

20 hours at most

Yes I do have a subscription to Colab Pro. I signed up for the $50 per month plan but would later downgrade to $10 per month plan because they had the feature where you can run a Colab book in the background but the Colab I used doesn't have that feature so I'd downgrade to the latter which allows me to use High-RAM to train quicker. No need to use the A100 GPU because I don't want to waste any GPU units that I bought.

nice

Thank you.

tough google doesn't allow use RVC again even though you using colab pro

Yes but you don't have to buy a subscription to Colab Pro. Google has a deal where you can "pay as you go" without paying for a Colab pro subscription. I've done that before.

he trained using the rvc v2 disconnected colab

me thinks

yes i am

Not really. I would've used Kit Lemonfoot's RVC v2 Disconnected Colab but it doesn't have the Tensorboard implemented in. I don't remember which Colab I used but it's the one Kalomaze made before Kit made their's.

the one with gradio

i get it now

Yes, that one.