Model Link: https://huggingface.co/VAP36/JackBlack/resolve/main/JackBlack_1860Epochs.zip
This model was trained off 16 minutes of super clean audio (Mixture of studio acapellas and carefully chosen songs)
Model works surprisingly well for speech AND singing, even though the dataset had barely any talking at all LMAO
and trust me this model isn't overtrained lmao, im surprised the Epoch count is that high too
#Jack Black - RVC 2 [TITAN PRE-TRAIN] [1860 EPOCHS]
1 messages · Page 1 of 1 (latest)
Rate the model
To rate this model, either use /rate or click the buttons below.
i dont think u where supposed to train the model to such a high amount lmao
300-500 epochs wouldve done it
ill send a 520 epoch demo of the same audio, ill let you be the judge
alr
still sounds good
just not as good
i know for a fact the model didnt overtrain though
and this was confirmed by multiple people that i asked
If you do want to use other weights then they are all here
https://huggingface.co/VAP36/JackBlack/tree/main
actually it does sound good
do you still have the tensor graph?
i mean the loss and gain ?
this was at 1670 epochs (my last screenshot)
the model works well for general speech as well as singing
i can't find the pth or index on your folder.
When did you know it was done training? I see that mel can keep going down if it wanted to
the /g/mel was flattening out but honestly it was still going down slightly however I kept making tests and there weren't any audible differences really, the 1670 epoch one sounds identical
And yk I was already near 2000 epochs so this took ages and I just wanted to finish it
I'm pretty happy with the result and I doubt it could sound cleaner
its there
downloaded it again to check, you can even see the (1) next to the file name
can't find (1)
have you extracted it
the (1) is just me showing that this is my 2nd copy because thats what happens when you have 2 of the same
meaning i downloaded it from here
@ornate shuttle
let me see
i did and it's weird
can you try to send the (1) links?
dude its LITERALLY this link
ive sent the link already
and its right there
for some reason its in a sub folder but it makes no difference at all
reupload for you, hope it works
https://huggingface.co/VAP36/JackBlack/resolve/main/JackBlack1860Epoch_reupload.zip
ok I see it
Best Jack Black model i've ever seen lol
I'M A MAN WITH ONE JACK MILE, SO MUCH TO DO IN ONE LEFT TIME, NOT MAN FOR A COMPROMISE NO WEARS A RISE, LIVIN' LAST, SO LIVIN' IT ALL

CLOSE ENOUGH MAN 🔥
thank you
You're welcome bro, it's insane that this was trained at 1.8k epochs and didn't overtrain lol
(Btw i have seen a TADC model trained at 2338 epochs if i don't remember wrong)
If anything it could be undertrained as I never actually saw the /g/mel go up I just saw it flatten
But to me it sounds perfect already
For me too
I may keep training it some other time but I doubt it'll make any difference at all
16m dataset too
I thought the longer the dataset, the higher the epoch count generally
Ig not
I always thought: more data = less epochs
Oh huh
Maybe irdk
The best models are generally created by actual fans of the artist or character
Like I doubt a non tenacious-d fan would know what songs to go for and stuff and would probably just pick one or two songs and call it a day
i like placing blocks and shit cuz im in fuckin minecraft
Model Ready ✅
This model has been synced with Weights and is ready to use for free!