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

I wanted to give everyone an update on this project!

I'm still working hard on this project and hope to have new models released along with a brand new GUI for v2 and v4 by the end of the month or sometime in November! I can't say for sure because this is a very taxing process computationally, storage wise, and mentally. I'll drop models between now and then exclusively for everyone here on Lite-records to test and play with.

Here's a list of what is being worked on:

- A new instrumental model with deeper vocal removals
- A new model that emphasizes vocal extractions for cleaner acapellas
- Brand new stacked models for both the instrumental & vocal model
- An overhaul to the GUI for v2 & v4
 
BJABox said:
That's a shame! I hope someone can replace it
Click to expand...

I can only guess but the thread existed because Anjok's models were not public, so he was taking requests for songs. Now that he has developed an app for it and has released the models publicly, you can convert any song you'd like by yourself.
 
djtayz said:
I can only guess but the thread existed because Anjok's models were not public, so he was taking requests for songs. Now that he has developed an app for it and has released the models publicly, you can convert any song you'd like by yourself.
Click to expand...

I will add that all the tracks from that thread, while really good conversions, were made with a different AI. That was the private one made by a different developer. It is a great AI, but the current best available AI in this topic is better overall. That old AI struggles with hard consonants, often creating these pinching treble sounds in the mix on them, and also really struggles with reverb. This AI is far superior on both fronts.
 
ChrisCall said:
I will add that all the tracks from that thread, while really good conversions, were made with a different AI. That was the private one made by a different developer. It is a great AI, but the current best available AI in this topic is better overall. That old AI struggles with hard consonants, often creating these pinching treble sounds in the mix on them, and also really struggles with reverb. This AI is far superior on both fronts.
Click to expand...

I agree. I also want to get in a live chat with you to go over how it is you get into the vocal layers better to get better background vocals. If it's something my computer can handle anyways.
 
ChrisCall said:
I will add that all the tracks from that thread, while really good conversions, were made with a different AI. That was the private one made by a different developer. It is a great AI, but the current best available AI in this topic is better overall. That old AI struggles with hard consonants, often creating these pinching treble sounds in the mix on them, and also really struggles with reverb. This AI is far superior on both fronts.
Click to expand...

This is very true. The model used to do those conversions was inferior to the "Multi-Genre Model 2" I released here a few weeks ago.
 
***UPDATE***

I've got a new model for everyone to try out! It's the first model made with the newest version of the Vocal Remover. It's NOT compatible with the GUI, as the GUI runs on v2. That means you'll have to run it through the command line. The GUI for v4 is almost done, I'm just working out a few bugs before I release it.

Here's the link - http://www.mediafire.com/file/9npipnmcytoffer/vocal-remover-v4.zip/file

To run a conversion, use the following commands -

Run on CPU:

python inference.py --input path/to/an/audio/file

Run on GPU:

python inference.py --input path/to/an/audio/file --gpu 0

If you receive an error it might be because you need to update Pytorch. You can update it using the following command:

pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html
 
djtayz said:
I've been trying to but I'm running into some errors, although Torch is already updated on Colab
Click to expand...
This may be unrelated to your errors completely, but one issue I ran into is that I can't get my GPU to run conversions on the cu101 updates. I am instead using the cu92 versions. I'm not 100% sure which ones have cu92, but for example, torch==1.5.0+cu92 or torch==1.5.1+cu92

You can do the same with the torchvision update if you like.

NewAgeRipper said:
I agree. I also want to get in a live chat with you to go over how it is you get into the vocal layers better to get better background vocals. If it's something my computer can handle anyways.
Click to expand...

I messaged you :)
 
Anjok said:
***UPDATE***

I've got a new model for everyone to try out! It's the first model made with the newest version of the Vocal Remover. It's NOT compatible with the GUI, as the GUI runs on v2. That means you'll have to run it through the command line. The GUI for v4 is almost done, I'm just working out a few bugs before I release it.

Here's the link - http://www.mediafire.com/file/9npipnmcytoffer/vocal-remover-v4.zip/file

To run a conversion, use the following commands -

Run on CPU:

python inference.py --input path/to/an/audio/file

Run on GPU:

python inference.py --input path/to/an/audio/file --gpu 0

If you receive an error it might be because you need to update Pytorch. You can update it using the following command:

pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html
Click to expand...

Could you elaborate a little better on this please? I replaced v2 with v4 in my documents. I added a flac file in the v4 folder of my Documents. But I'm not quite getting anything out of the cpu command. Could you maybe lay out the command as though you have an actual one ready to go please for us dummies here?
 
ChrisCall said:
This may be unrelated to your errors completely, but one issue I ran into is that I can't get my GPU to run conversions on the cu101 updates. I am instead using the cu92 versions. I'm not 100% sure which ones have cu92, but for example, torch==1.5.0+cu92 or torch==1.5.1+cu92

You can do the same with the torchvision update if you like.



I messaged you :)
Click to expand...

Using CMD, I also was unable to run the new model with GPU, but it ran fine with CPU
 
djtayz said:
Using CMD, I also was unable to run the new model with GPU, but it ran fine with CPU
Click to expand...

That's the issue I had. rolling back to any +cu92 version worked for me
eg: pip install torch==1.5.0+cu92 torchvision==0.6.0+cu92 -f

Or if there's a more recent cu92 available then
pip install torch==1.6.0+cu92 torchvision==0.7.0+cu92 -f

If this doesn't get the GPU conversions back and running then I don't know. That's just how I fixed it for me.
 
Sorry for all the confusion guys! I might have to overhaul the whole instruction set in the main thread and do more VM tests. This is usually why I wait to put out new releases until I've worked out the kinks at every angle, which I'm still doing. I only released this v4 model because it's currently my best performing model and I wanted to give everyone a chance to test and play with it until my final releases were complete.

In the meantime, post your errors here and I'll help as much as I can.
 
You must log in or register to reply here.