Description
This model goes across almost every era that has stems available besides from kid Michael. (Uses Off The Wall Thriller Bad Dangerous History and Invincible.) 2-hour and 16-minute dataset. Dataset by me Trained by
Comments
Michael Jackson (RVC) 250 Epochs [Every Era] (no child Michael)
training ?
or how long it took to get the dataset
I'm pretty sure it was under 6 hours
maybe 4
it is trained to the max what I've been told
no overtraining
ik but better than yt links
This model is exactly what I was trying to attempt until I found out the stems had headphone bleed 😭😭
But amazing work though
Oh yea they all isolate out headphone bleed
only main lead vocals didn't bother adding harmonies or doubles to the dataset
How did you remove the headphone bleed exactly?
If I know how I’ll make a v3 of my MJ model
basic noise removal does this but the clapping and stomping stays most the time which i love but bad for datasets 😦
What about MJ clicking his fingers?
Just basic isolation
I think it was kim vocal 1 or smth
he does that and stomps and everything on some it’s wild you heard the 2 bad raw vocals?
I made the dataset ages ago
back when diff-svc was still a thing
For the stomps If they are super loud or not isolated enough I just eq them out
Thank you all
def the most annoying part was isolating each vocal.
I’d say it took me like a few days to go across all stems
Did you use the ones in the remixpacks archive or the MJ-stem txt?
I just got them from a Facebook group
which was the mj-stem txt pretty much
If someone ever wants to go through the suffering and pain of isolating the doubles and harmonies gl.
they’ll be good to the dataset maybe an extra hour to it alone
Thriller starlight session and off the wall multitracks have very clean individual harmonies and I used them on my v2 Michael Jackson model
i don’t remember which songs had a lot of harmonies in one file
probably ghosts
one thing it doesn’t include is Jackson tour vocals or whatever
Obviously not because there’s no HQ vocals for them
ah
I also didn’t add little Susie cause at the time there was no deverbing method that was as good as todays
Can you release the edited dataset that you used?
he spent days making it so prob no I guess
Not atm
ima hold onto it for a little while
⚠️ I will be accepting a few commissions DM me and we can talk about prices ^This also goes for mixing AI covers for you‼️ ⚠️
this sounds just like him
What about this is it and Michael era?
Add a comment
Samples
1. Singing
Male
English
2. Singing
Female
English
3. Singing (Dry)
Female
English
4. Singing (High)
Female
English
5. Singing 2
Male
English
6. Singing (Dry)
Male
English
7. Singing (Dry, High)
Male
English
Pitch
Weekly Metrics
Users also tried
More to explore
Loading more