I searched on discord but found no one with the same issue.
I first tried 1.8.1 version and encountered same error.
I juste wipe everthing and reinstalled 1.9.1, still the exact same thing, stem separation causes errors each time.
Will wait until new version is out.

 

Last edited: Nov 14, 2023

maybe - voice repainter or voice repaint

 

I made it myself, which will be part of the pro side of Weights soon!

The thing with this tech is If you have high-quality audio when you make the model then the output will be high quality.

Make your model with a Neve Mic Pre, and a LA2A It will keep those characteristics in the model.

 

The assertion that compression is 'ruining music' is entirely subjective. Compression can be utilized to craft a new ambiance, similar to how distortion was once viewed as 'ruining music'. When electric guitars were first distorted, many initially perceived the sound as unpleasant. However, this technique eventually elevated rock music to a new level. As the quote from 'Back to the Future' goes, 'Your kids are going to love it.'

Regarding my concerns about quality, it's important to note that genres like early punk, SoundCloud rap, hyperpop, and countless others may seem to have 'poor' quality, but this characteristic contributes to their unique appeal. What one person deems bad, another may consider a sub-genre.

If you're seeking a high-quality model with replay capabilities, it's feasible to create one using high-quality vocal data. This is in contrast to the models developed by the community, which often rely on ripped acapella tracks.

As for the difference between 'Repaint' and 'Replay', I assume that the 'Paint' aspect of 'Repaint' suggests some form of visual art integration. Meanwhile, 'Replay' connotes fun, as well as the action of pressing the play button during the creation and enjoyment of music. It's like saying, 'Let's hit play and listen to that again.'"

 

can you explain more in depth why you think I should change the name of the app? opposed to what It is already named?

 

We are working on a bug fix.. Hold tight please

 

Someone could put in a RAR pack file on Cloud the best 20 english singers voices ranking, male and female, to be easy for everyone to download.

Anyone who has it could help all of us...!

thanks
and all the best

 

We are putting a voting system and playlist curation on weights.gg to help curate out the bad and feature the good!

 

Last edited: Nov 13, 2023

Can you place the current version number on the download page so we can see what we are downloading?

 

Hi man
This needs a more detailed explanation than this for offline training than the Git repository provides not just Windows
A detailed break down of install instructions would benifit everyone

So for now until this detailed tutorial is provided anyone who wants to train datasets use the Rvc Software
As it installs everything
Replay will load the output trained vocal dataset by drag and dropping into the application so it installs
Then you can utilise your own trained vocalists

 

@curtified I have went to discord none of these are working for me at all I started with I think 1.8 or close to it up to the last one a day or 2 ago the 1.9.1 or 1.9.2 none of them let me do anything, as soon as I click to start it at the bottom it shows the percentage then I get an error it even does it with the default talking clip.I was asked to share logs which I did I even tagged "musicman" or some name like that who told me to share them and I have yet to hear anything back.I really want to try this out but right now it's useless to me im on windows 11 I assume on the latest update but if not I'm not far behind.Every update I did it was atleast 3 or 4 all had the same outcome.

 

For those interested in taking this offline after they download a few models:

I manually downloaded via the app download manager all the models available there. this numbers don't include the ones that glitched. I ended up with 3502 models which sum up to ~450gb; Most of them are between 50mb and 200mb, some are around 400 to 500 mb and the biggest one I found was "Low G" at 1.86gb

I tried to make a torrent file so i can share the whole collection but something went wrong and I don't have time to investigate (especially after this endeavor lol)

@curtified Some improvements that this software would really benefit from (especially for large offline collections) would be:

- adding tags on the models
- the ability to change the directory for exports and downloads so users can do that only on their burner SSDs
- removing models named "Unknown" or the ability to rename the models once their on the local storage

Cant wait to get some free time and explore more; I played with it for about an hour before starting to download all of those gb and I didn't expect to have so much fun. Never had a crash till this point (not even the one with the stems thing).

img for refference for "glitching" while downloading

 

Last edited: Nov 14, 2023

Thank you for all the feedback and suggestions. We will definitely incorporate some of these features into the app.

I'm glad you're having so much fun! That's exactly what we aim for. Whether it's for professional use or just for fun, this app is a great tool to experiment with!

Regarding the issue of the models being large and poorly named, as of now, all models are community-created. Sometimes, people include too many files in the zip archives they create. They also often don't rename the files, so you might see them labeled as 'Unknown.' We are planning to add renaming and organization features to the app soon, so you won't have to manage these files manually with your local file system.

 

Ugh, it bums me out to read all of this. I'm really sorry you're dealing with so many issues. When developing apps, the only way to test them on a wide range of machines is often through public failures.

My goal is to resolve these issues so that you can use the software effectively. Please know that we are working on it and your message has been noted!

Have you tried turning off the 'stem mode' in the settings? That has been a common issue for many users.

 

hey hey! The git I provided is just the current training get we suggest to use. I have no affiliation with it.

We hope to add training to Replay soon! until then we have to go 3rd party for model making. I'm sorry if its a little frustrating.

 

Congratulations on your application and storage site
sounds great that the training will be included in the future updates of replay
that will make the application all inclusive ... well done
at present the replay application your developing works very well Ive had No issues here
With replay

On the Amd Ryzen 5800x with 32 gig and running Geforce Nvidia Gt 1080 8gb using the gpu it rocks and completes a change vocal full track in around 60 seconds
Cant complain about that

I also train NaM profiles for Neural amp modeler offline in anaconda with python that take around 4 minutes at 400 epochs Each

Utlimate vocal remover can split a demucs 4 or 6 track in approx 4 minutes as well

Thats why i asked did you have the details to setup offline training for datasets

Anyway ill wait for the training in replay to arrive keep up the good work there


Cheers

 

Last edited: Nov 14, 2023

Pm me and maybe I can help with training

 

UPDATE for my PC only users with training:
Feel free yo use this GIT to train your models.. https://github.com/IAHispano/Applio-RVC-Fork there is a .bat that should make it easy to run.

 

@curtified I found the explanation here as the issue is similar:
https://github.com/lmcinnes/umap/issues/702#issuecomment-1002396093

This is due to my AMD CPU, it seems that your software wants to use an intel only library "svml_dispmd.dll" which is located on my system in C:\Program Files (x86)\Common Files\Intel\Shared Files\cpp\Bin\Intel64

I renamed the file to svml_dispmd.dll.bak and restarted Replay and now it works!

This is a dirty fix, so I hope you'll fix this in the software by a proper CPU detection or allow end user to use / not use this library in settings.

I posted this workaround in my discord issue as well.

 

Last edited: Nov 14, 2023

For the Mac ARM version, I'm curious why there is no "open in rosetta mode" on the app's info pane? Is it not an actual ARM app?