Yeah if sound quality is an issue. I would suggest using UVR in ensemble mode to get the best version of extracted stems then pitch those in your daw. It takes some playing with your model and the pitch to find what vocal range sounds best with what pitch the dataset is made for

 

well i think its been very well proven as a techology that can be useful when applied in the right process
My time to showcase is done time for others to pick up the batton and continue the journey
its been a blast
thank you for creating your application
have a great christmas
cheers
Rodger

 

it's getting super hilarious

 

Last edited: Dec 20, 2023

Ariana singing+rapping Riders On The Storm (from NFS UG2)

 

what do you mean exactly? UVR Karaoke stem method? or is there instrumental-only mode in Replay.AI app?

 

finally getting gist of it, Avril Lavigne singing Fallen Leaves by Billy Talent is almost natural perfection

 

Karol-G singing Sweet Child O'Mine

 

.updated

 

Last edited: Dec 23, 2023

updated

 

Last edited: Dec 30, 2023

So we put the algos of UVR into replay. but not in ensemble mode (yet) it takes a lot more time and compute to run those algos. If you want a better sounding acapella consider downloading UVR stand alone and making stems with that.

All the information is here on this thread I started a while ago:
https://audiosex.pro/threads/uvr5-the-best-ai-stem-separation-algo.69463/

 

throughout the years i've collected a whole bunch of samples that i was never able to use (clean up issues, quality issues and such) and this made it possible to finally use them. this is really awesome dude

 

So happy to see this! One of the many uses and growing for the replay tool we created!

 

@curtified i have spend like forever to try to figure out how to download the RVC from your google research, how on earth do I do that and will it end up in a index and a pth file or do I have to convert it afterwards?

I am also interested in your Cinematic Voiceover clone. Is it possible to pre purchase before the weights pro arrives?

I hope you will implement a clone option inside Replay soon, I really dislike using the Google Dev. tool and restarting everything all the time . Keep up the good work! Love what you do!

 

in the side bar on google colab there is a folder called RVC-webui in there is the there will be a folder called models/checkpoints.

you can also use replicate to train. https://replicate.com/replicate/train-rvc-model

as for in app model training. It takes a lot of gpu using CUDA. So only a narrow window of users machines are compatible.

 

Wow thank you for this. Other engines like this are charging way too much. This is great for the community and thanks again.

 

Youre welcome! they are just charging for the compute to use the free tech.

we just made it so you can use your local compute easily.

 

I am stepping away from AS for a while due to false accusations against me in this thread and others. Feel free to review my past history of posts, which show that I have always had good intentions.

Being involved with a sister site to a grey market software source for nearly a decade, I've observed that the negativity on AS, especially towards me, often reflects others' inner hurt, anger, and fear. This is understandable in a community striving to make a living through music, where like-minded individuals might unintentionally create an echo chamber for such emotions. Nevertheless, I have always found AS and AZ valuable for discovering new and underground tools and information in the music and tech space.

Despite any criticism, I hope my roles in Replay, Rubin, Flosstradamus, Suno, and other projects are seen as positive contributions to AI and music. However, given the current climate, I am stepping away from AS. For any issues related to Replay, please contact us on our Discord channel.

 

Last edited: Dec 27, 2023

RE the use of: https://replicate.com/replicate/train-rvc-model

I can't set it past 100 or so Epochs. The voice training process seems to 'fail' at around the 1 hour mark (or at around 135 epochs). Is this intentional? Is there a way (by paying for time/use of the GPU's) to increase epochs and thereby increase the quality of the RVC models you're training?

Any help/guidance on this would be greatly appreciated. Thanks!

 

Sorry to hear mate,best wishes for your future ventures Hope to see you return soon
few more added to the playlist

Marie Fredriksson Aka Roxette trained vocal dataset
T-T-T-Take It

Marie Fredriksson Aka Roxette trained vocal dataset
Hold On My Heart

 

I couldn't get it to work either. I ended up with buying developer credits on Google because free didn't work (around 11€ for 50 hours of render). With the tips from @curtified it was possible to find where to download it . Highly recommended! Sounds amazing!