Hi everyone, first post here, I'm fairly new to this, so I was wondering if anyone could share tips on how to process vocals after using UVR models. I'm trying to get rid of those weird, annoying artifacts that sometimes remain.

Most YouTube tutorials focus on processing vocals recorded directly in a studio setup, but I rarely see advice on cleaning vocals that have been separated or already processed. Any help?

 

what algo are you using in UVR? the best for UVR is ensamble.

the multi stem algo in suno is the best ive heard. The multi stem algo actually regenerates the audio using their generation engine. Less artifacts when its generated.

 

i'm not familiar with suno. i looked at their website, and i am wondering: can i upload a song that my band recorded back in the 1990s on a 16-track tape demo, and use suno to split the stems? i don't have access to the individual tracks any more and i'd like to see if i can do a better job of mixing the demo today with all the new tools that are available.

 

I checked this document that I've found on the forum "https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/edit?tab=t.0", checked vocals and the ensemble page but can't seem to find those models

 

Yes you can! You can also PM it to me and ill do it for you.

 

They are all downloadable within UVR app itself. Its been a while since ive used it but you have to go into each model type and download the various models.

https://audiosex.pro/threads/uvr5-the-best-ai-stem-separation-algo.69463/

 

Good question. This probably isn’t the answer you’re looking for, because it’s not graceful, but I usually end up massively distorting extracted vocals which masks artifacts. That or drowning them in reverb.

I am often isolating speech from films and stuff like that. Slapping something like izotope VEA on extracted speech might be a quick way to thicken them up.

 

When i have to do that, i tend to do like ddpp suggested, and add more stuff on it to mask artifacts, or make it seems like they come from your processing.

But before hand, just do a very extensive EQ on it, you have to consider that the file you're working with is "broken", it aint a clean u87 vocal that you would treat with a +1 db on some frequency, it's a weird sounding voice that you're trying to save. So weird ass looking EQs, Soothe can help getting rid of some harsh frequencies. One trick i wouldl add is to add layers that might get the vocal momre "full".A lot of time with AI extraction, you will loose infos in some frequencies (and sometime will vary with whats happening around on the song), so record a backing vocal that you will layer with it, to recuperate some of those infos, or try to play the melody with another instrument and layer it.