|
--- |
|
license: cc-by-nc-sa-4.0 |
|
language: |
|
- en |
|
tags: |
|
- wubby |
|
- paymoneywubby |
|
- twitch |
|
- streamer |
|
- regards |
|
- so-vits-svc-fork |
|
--- |
|
|
|
Our favoutie streamer needs a better singing voice - so lets work on that together. |
|
|
|
I formed the model off of the xmas 2023 Wubby Sleep Aid from Youtube. Dataset on my profile. |
|
|
|
I tore it down, isolated vocals, split to over 1000 9s audio clips. |
|
|
|
Training initially was started on the following system: |
|
|
|
* Windows 11 - direct, not using WSL |
|
* i7 11700k with Lian Li Galahad 360, running at 100% |
|
* 1x 3070ti 8GB with 32GB DDR5 System RAM |
|
|
|
After 100 epochs, I moved the model over to my AI training rig: |
|
|
|
* Windows 11 - direct, not using WSL |
|
* i9 13900k with custom loop cooling system running isopropyl alcohol |
|
* 2x 4090ti 24gb each, with 128GB DDR5 System RAM, all liquid cooled with above system. |
|
|
|
After 250, uploaded, but am continuing to train and will upload checkpoints as it goes. |
|
|
|
Contact me on the PMW Discord in the #nerds channel, or ScottishHaze directly (note, my name on the server may change to match holidays, currently it's "NewYearsHaze", just look for "haze" and it's prob. me ;) |
