Okay, as promised, I wanted to share with the AS community the FREE app I've been working on for voice and instrument cloning.

https://www.tryreplay.io/ It runs locally! No server is needed! We even have stem separation, de-reverb built in. We're about to launch a plugin in a few weeks too!

We've also built an AI model hosting platform over at https://www.weights.gg/ (with 75k+ users & 16k+ models). Think of it as a Splice for AI models. Please download or contribute your models to our library!

We're open to feedback from the music-making community, so please share any thoughts here or on our Discord(s).

 

I noticed on other posts people have been asking me how to make AI models for this software.

We will add a training feature very soon!

Until then you can use this google colab:
https://colab.research.google.com/drive/1DqRzmO-TO2z-DsJyzShfAysUvPxz-EdX?usp=sharing

If you have a PC and know how to run python code via command line you can just use the git hub repo:
https://github.com/ddPn08/rvc-webui

 

Last edited: Nov 13, 2023

I have tried to install replay on two windows 10 machines. The install is stuck in a loop "File 18 of 18" and repeatedly says download. When I click download it goes back to "File 18 of 18" I dont have time to troubleshoot. best of luck.

 

Looks cool. Will keep and eye on this, but likely wait for the plugin.

Once piece of feedback for the weights website is that only one preview should be able to play at once.

While listening to a preview users will probably want to keep scrolling for other models and when clicking on another one the other(s) continue to play.

Clicking a new preview should stop the currently playing one as to not have to scroll back to manually stop it. 16k models is a lot to browse, so staying at the location on the page to manually stop previews before moving on is quite a bit slower.

 

Wow thank you for the feedback! Will get all of these implemented ASAP!

 

update to the newest version. 1.8.1 should fix this issue

 

I just tested...
It is pretty darn good.

I was wanting to test the instrumental aspect and only saw the vocal changing.
But that part, even, performed really well. And the interface is very user friendly.

 

I appreciate that! all feedback is welcome!

There are a few instruments we have on weights.gg.. you can also train your own using the links above.

 

Downloaded it. It needed models and downloaded 5 gb of data and then a 2.5 gb again but stuck at file 16/18 for 20 mins. I shut down the program and uninstalled it but the program was 400 mb. Where did it download the 7 gb data?

 

C:\Users\"YOUR NAME"\AppData\Roaming\Replay\com.replay.Replay

 

I'm intrigued!

Installed, and am downloading the AI Data...
... But something's gotta be done about the download speeds. I think my 14.4 baud rate modem from 1994 downloaded things faster.



EDIT:

"Depending on your internet connection, this might take a while." lol

I have gigabit service & WiFi 6, and to give context, I've downloaded 75 megabytes over the past 1 hour. (2%)

If my math is correct, it's going to take 50 hours to download the 3.13 GB.

Sorry, but I'm going to have to pass.


EDIT 2:

wow, okay! Just before I went to cancel the operation, the download speeds went from snail slow to blazing, rocket ship fast! The download is almost finished. . Will give it a whirl.


EDIT 3: I couldn't create "songs" or outputs due to relentless server issues. Bug reports were created, but I haven't the time to hop on to discord to report them. Perhaps another time...

 

Last edited: Nov 9, 2023

Man you are so intelligent, this is absolutely amazing! I would really love to see a tutorial on how to train new voices or instruments. So far everything works smoothly here, one little thing, I saw the final output is *.mp3, is there a way to change that to *.flac, also, is there a GPU processing option?

 

Last edited: Nov 9, 2023

hey! i tried the app today but i get the error "no valid pth files found for this model" on most of the voices that i tried downloading through the app :/
EDIT: also the index file is missing in the folders of those models giving out that error

 

Last edited: Nov 9, 2023

This is damn good! Thank you for your great work.
If a learning function is added without a web connection, it will be a masterpiece!
ps. Google collab training failed(

 

Everyone is training for the marathon, joke aside google is probably overloaded with all the ambitious trainers. I would suggest to be patient, this is all new and revolutionary for the most of us.

 

I will fix this bug.. I don't know what happened. Let me look into it.

 

Let me see on that colab.

The learning (Traning) function only works for PC right now. It uses a PC native CUDA algo to train the models.

Soon tho!

 

Hey no worries. We made this free for the community. Sorry Its frustrating you. Were just trying to solve a problem for the community and learn in the process.

 

Nice one, @curtified
Working great here, had couple of hiccups for downloading model, but restarting program just continued where it left off, which is really great.
Right now I'm using it on my AMD laptop, which is slow due to integrated graphics not supported. Can't wait to use it on my desktop.
Keep up the good work!!

P.S. Post updates on change log in this thread, if you can and have time

 

I appreciate you! Im excited to see what you notice on a machine with GPU. Just a note. I’ve found I keep my audio length shorter for cpu renders. Things under a minute seem to go pretty quickly on cpu.

I will keep this thread alive with feature for sure!