So-Vits-SVC Vs. RVC: A.I. Vocal Comparison

Kanye going hard on AI right there 😂

Great video!
Although in my testing so-vits seem to have slightly better output (cleaner, less static), I prefer RVC, due to its fast re-transposing after first run (cache).

Can train on 3070ti so-vits 8GB. It does takes a little longer though.

thank you very much man but can you explain in deep about training in rvc v2

I've been on the fence about actually trying things out myself. I have SD running, but sovits was a bit daunting. This video has convinced me to try rvc 💪🙏

kindly make a tutorial on how to train on DDSP SVC....you forgot to make that after you promised in one of your previous video

how big should the dataset be band how many epoch to make a decent model, in your experience?

Does So-vits-SVC do better with cloning a voice with emotions such as Laughter? Can't seem to change laughing voices without it sounding "ta ta ta ta"

do you have any idea when will rvc v3 come out

RVC, FTW!! 🏆 IMO, RVC sounds waaay clearer and could become the preferred sooner than later 🔥👍

Good video, but I think you are wrong about the so-vits-svc system requirements. I am able to train 500 epoch models with 25-30 minutes of data in about 10 hours using my GTX 1060 6GB, batch size at 5

Both are still pretty good, just sounds like RVC is 1080 and Sovits is 720p

Sovits runs on linux with no problem, also using amd gpus (with rocm driver). In fact, you can't use amd gpus with so-vits on windows, you need Linux to use rocm drivers.

Please answer my question, do you have Telegram or Facebook to communicate with you? I trained an audio clip and reduced it to 10 seconds and trained it. Do you know how long it takes for my laptop Lenovo core i5 10th generation RAM 12 and graphics card amd Does it succeed in training Please answer my question

I'm using RVC.
And at the moment the most important thing is not to choose between Vits and RVC

I think it's a data set and how to record your own voice to drive (even the sound of your breath if it's a condenser microphone). I'm using iztope and trying different methods. do you have any special tips?

Thanks! I've been digging up whole Internet for all the guides and best practices for so-vits and now I feel like RVC's version at the end of the video sounds better... I feel a bit upset but I'll stick to so-vits anyway - now that I'm familiar with it good enough

rvc is nice but somewhat robotic, in that so-vits had more artifacts but also had some different tonality to it, especially when you would do multiple takes on the same vocal you would get slightly different outtakes!

Btw, google colab runs on linux, so so vits definitely does too

Unlike SVC, RVC seems to handle pitch better and it's also easier to train or set up.

I love how you try to cover all the possible points and issues that may happen. You have earned a new subscriber. Keep up the good work💪🏻😁

You said you gone leave some places to get models from linked below, but i cannot see these places in your description?

both sound like shit when converting my mildly deep voice(nothing outragous or unusal) .. have to speak differently for the ais to make sense of it. woman to man works great though.

feels like so vits has less punch

Fantastic look under the hood of some new magic. Having done mashups since 1998... for me, it's overwhelming and fun to see the breakthroughs that enable creative people to take ideas and make them happen. I've never felt that time learning tech should become a moat. Just because I've sunk two decades into hand-crafting a result... doesn't mean I should resent it becoming "easy"... I welcome all tools and "cheat codes"... Let's go!!

Yo p3tro! Have you gotten into Chirp or Bark models? Would love to hear your take and explore running locally.

I'm glad you went over the pre-rescue-its so thoroughly🤣 But seriously, good video, very helpful!

Hi P3tro, I've been getting into RVC and I noticed that it really struggles with converting pure "mmh" and "uuu" sounds, even if very clear samples exist in the dataset.
I think its the retrieval component that fails to recognize them.
Have you had this issue and found something to improve it?

which one does real time conversion?

Can you have both installed?

I know it was a year ago, but the "Chinese" is actually Japanese XD
other than that, Very cool information!

Totaly RVC Main
I'm French so I don't hear the flaws. For me is more natural.