What's new
By:
Filters
LiteRECORDS

Register a free account today to become a member! Once signed in, you'll be able to participate on this site by adding your own topics and posts, as well as connect with other members through your own private inbox!

  • Guest, before your account can be reviewed you must click the activation link sent to your email account. Please ensure you check your junk folders.
    If you do not see the link after 24 hours please open a support ticket.

Powerful New Vocal Remover AI - Instructions

Danizm said:
Get this man some studio acapellas!!! how many would you need?
Click to expand...

All he has to do is get with Brad about access to all acapellas on site for any he already doesn't and ask for those that are not here. Brad has said he would like to get behind this and support it. But when things go quiet, it doesn't really make it look promising enough for Brad to get behind. Silence kills a project. I could give him every accapella there is for Rock Band 1-3. All the Jammit pellas. Guitar Hero pellas. There's over 3,000 he can use to train with. Not to mention pellas from other sources.
 
Last edited:
NewAgeRipper said:
All he has to do is get with Brad about access to all acapellas on site for any he already doesn't and ask for those that are not here. Brad has said he would like to get behind this and support it. But when things go quiet, it doesn't really make it look promising enough for Brad to get behind. Silence kills a project. I could give him every accapella there is for Rock Band 1-3. All the Jammit pellas. Guitar Hero pellas. There's over 3,000 he can use to train with. Not to mention pellas from other sources.
Click to expand...

For an effective model (at least as effective as the multi-genre model), it will take at least 400 pairs. They also must align properly with the official mix. If they're from stems, then the pairs can be made from them. Putting them together takes A LOT of time. However, if enough are sent to me as confirmed pairs, I will start training with them immediately as it will save me a lot of time.
 
Anjok said:
For an effective model (at least as effective as the multi-genre model), it will take at least 400 pairs. They also must align properly with the official mix. If they're from stems, then the pairs can be made from them. Putting them together takes A LOT of time. However, if enough are sent to me as confirmed pairs, I will start training with them immediately as it will save me a lot of time.
Click to expand...

I sent you a private message.
 
***UPDATE***

So, after some testing I've figured out a way to get cleaner vocal removals! I will be releasing a new model for v2.2.0 as soon as next week. It will be another multi-genre model. One for v3.0 will follow as well.

With that being said, I think it's important that we treat v2.2.0 and v3.0 as separate AI's because they both have weaknesses and strengths over each other.

v2.2.0 - Vocal removals are a bit tougher, but the instrumental quality suffers little to no degredation.

v3.0 - Vocal removals are more aggressive, which is good. However, it comes at the cost of higher instrumental quality degradation (think PhonicMind conversions).

Last thing, if anyone has pairs they would like for me to include in my next training set, please follow these instructions before you send them -

1. Please use stems only. You can get them from Rock Band, here on Lite-records, or other places
2. Use those stems to create a mix, an instrumental, and an acapella
3. Name the files with the artist and song name followed by "_mix", "instrumental", and "_voc" (Exp. WEEZERSAYITAINTSO_mix, WEEZERSAYITAINTSO_inst, WEEZERSAYITAINTSO_voc)
4. Make sure you do this all in the same session so they will align by default.
5. Please make sure the instrumentals are clean of background vocals. Also make sure the acapellas are totally clean with absolutely no instrumentation bleed through.
6. Send them using Dropbox, drive, mega, etc
 
Has anyone tried to use a converted instrumental with this to do a phase inversion and get the vocals? I'm guessing it will not work but had to ask anyway. :)
 
INGSOC said:
Has anyone tried to use a converted instrumental with this to do a phase inversion and get the vocals? I'm guessing it will not work but had to ask anyway. :)
Click to expand...

This AI gives you an instrumental and acapella tracks. The acapella one is basically what you described - it phase inverts the instrumental with the original, you won't get a different result if you do the same thing yourself.
 
INGSOC said:
How hard would it be to make a standalone version? Something to just drag and drop files into?
Click to expand...

They have tried that route and just can't get an exe to build properly. For now there is Google Colab method and the Python files GUI method.
 
INGSOC said:
Doesn't work on W7 :(
Get this messageView attachment 674
Click to expand...

Run this -

pip install opencv-python

I've tested my instructions on fresh installs of Windows 7, Windows 10, and Ubuntu Linux 18.04 with no issues. Please let me know if you run into any more issues or if that works. Also, NewAgeRipper is right, I tried to have this compiled as an .exe but can't due to known issues with PyTorch. I even brought in some really smart people in to help with that but it's just not possible yet.
 
***New Model Update***
9/1/2020 - Please see my GitHub release here

These models are pretty big improvements to my previous ones. I'm working on training more in the coming weeks so expect more updates in the coming month.

Enjoy!
 
Anjok said:
Run this -

pip install opencv-python

I've tested my instructions on fresh installs of Windows 7, Windows 10, and Ubuntu Linux 18.04 with no issues. Please let me know if you run into any more issues or if that works. Also, NewAgeRipper is right, I tried to have this compiled as an .exe but can't due to known issues with PyTorch. I even brought in some really smart people in to help with that but it's just not possible yet.
Click to expand...

Getting this error. Any ideas? thx
118684762_639293123638489_8336882866219162968_n.jpg
 
You must log in or register to reply here.