Paul McCartney (1965-1980)

Old Mangio Crepe Model: (36 min) https://huggingface.co/Albinator/PaulMccartneyYoungV2/resolve/main/PaulYoungV2_e340_s14620.zip New Mangio Crepe Model: (44 min) https://huggingface.co/Albinator/PaulMccartneyYoungV2/resolve/main/PaulYoungV2.1_e370_s42550.zip (Use Kim Vocal 1 on it or something else <:catblush:1146690213679202304> ) Alternative version below in the repo (Epoch 440) . RVMPE Coming soon. Info: This is pretty much the same model as https://discord.com/channels/1089076875999072296/1104051594552688803 but with 8 min of extra samples and trained on Mangio-Crepe on V2 and RVMPE 48khz, 8 Batch Size. The link to the best versions imo are above but here are other versions with more or less epochs: https://huggingface.co/Albinator/PaulMccartneyYoungV2/tree/main My other Paul models and his 1st wife Linda: https://discord.com/channels/1089076875999072296/1123952175689777182 https://discord.com/channels/1089076875999072296/1123943860452720750 https://discord.com/channels/1089076875999072296/1125564138941317181

what albums did you use in your dataset?

these are the files that i used

Sgt Pepper RRS White Album Venus Revolver Let It Be London Town Abbey Road

there's more

I have tried pretty much everything, there is a lot of double tracked vocals and some stuff just don't sound good.

I have more samples that i thought sounded good but i didn't like how the model sounded so i removed them. Ill take a second look at my extra files but right now i really need a break.

1965-1982

yep, gonna make a model that's in the 90's soon too

you should do like

a really bad paul model

where the dataset is just Bip Bop

it'll be funny

(for the few people who get the inside joke)

😆

Paul McCartney (Young) (RVC V2) 560 Epochs

Paul McCartney (Young) (RVC V2) 340 Epochs

I'm gonna make a RMVPE version and also redo everything since the dataset is old af and was just trained on mp3 to wav files <:nails:1091517541240807424>. It will be finished next month. I also have to say that VOC-FT separation is so fucking good that it makes remaking this one even more worth it.

trained on mp3 <:skull1:1093358663151923301>

The dataset was made back in April when all the AI shit was exploding, and i didn't know that much about what to use and i was a bit lazy. But i basically have blueprint on what songs are good to use so its going to be a lot eaiser to make, but i still wish i had made it in lossless tho. <:skullflushed:1093200019508699136>

I'm training a RMVPE and an updated Crepe version now. Its gonna take a bit longer since i need to re-train since colab fucked everything up <:calmcryandsmokin:1093342214765150399>

Paul McCartney (1965-1982) (RVC V2) 340 Epochs

A new Mangio-Crepe model is up now and i think it is for the most part better but worse in some circumstances (Epoch 440 has slightly better sibilance but his voice sounds deeper). I have trained RVMPE up to 300 epoch but it sounds like shit atm so i need to train a bit more. I might also retrain it with Kim Vocal applied to it got get rid of some stuff. I might also post more samples soon.

RVMPE is up now 🙂

Paul McCartney (1965-1982) (RVC V2 | RMVPE + Crepe) 340 Epochs

Exelent!

Updated model: https://www.weights.com/models/clpobeghp000cpbtfzzkblk7f