What's the best model for extracting instrumental only? #444
|
What's the best model for extracting instrumental only? |
Replies: 14 comments 37 replies
|
MDX-NET UVR-MDX-NET Inst 3 but if you want to get highest quality result you should try using ensemble mode instead (combine results of multiple models together) |
|
Do you know for sure that it will finish properly if I buy some credits? I can't figure if it means uninterrupted work or it can still throw a captcha mid-process. Or I should buy a VM from them and work on it. The latter is too technical for me though, haha. On the other hand, don't wanna throw money in vain. |
|
id suggest just using UVR then :) I think if you input the VIP code in the model download center there's a variant of the MDX23C model there to download. |
|
Oh, I know, but it takes way too long even on my $6500 Macbook Pro M2 Max. I'd rather figure out how to use it on Google Colabs :) |
|
Please suggest models to only extract music from the track no drums, no vocals |
|
How many more models I need to download if I use "Ensemble Mode" ? |
|
For the cleanest possible results for now that pretty much removes all of the vocals with little to no vocal bleed, I use MDX-NET: UVR MDX-NET INST 3 + UVR-MDX-NET Inst Main + 7_HP2-UVR +Demucs: v4 | htdemucs_ft - Shifts: 2 - Ens. Algo.: Avg/Avg Denoise: Off |
|
How do i choose all those models? I can find UVR MDX-NET INST 3 and Demucs: v4 | htdemucs_ft. But how do i choose the others? Is it inside the secondary model settings where i can choose a model for vocals, other, bass and drums? |
|
which type of your Laptop level?? RTX4090?? |
|
Mate I am on a 5070 and it took 8 hours to filter a 4 minute track, what are you on |
|
Damn...I am on a core i5 and intel uhd graphics and a 4 min track takes around 3 mins. What are you doing? |
|
Somebody please tell me what "Shifts: 2" means here? |
|
im doing voice overs of a certain show and i want to extract the best instrumental part as possible! im having trouble finding the best ones do u have any ideas by chance? |
|
I just need to separate the main instrument of the instrumental track. ( for instance, the lead instruments like guitar, or accordian or harmonica or piano from the instrumental track. Please suggest the Model/s and the processing method. Thank you. |
|
Thank you very much for your explanation. I will try to figure out where in the app i can create this ensemble processing chain
Il 11 ott 2023, 04:13 +0200, Lyssa ***@***.***>, ha scritto:
… > Hi there,
> sorry for my stupid question but I am new here and trying to figure out how it works. When you list the modes you use in your ensemble workflow, you mean that you process the output with a new mode every time, or did you get to apply multiple modes at the same time ?
> thanks
there's an "ensemble mode" and you can select multiple models to run your song through. nowadays I strong suggest you to use the new MDX23C model and place it first in the ensemble (you can download it in the Download Center in UVR)
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
|
Hi everybody, |
|
Este programa é simplismente fantastico, o melhor de todos os tempos, parabens ao criador... |
|
Personally I'd say just download UVR 5 and run it locally; even if you
don't have a good GPU it can run in CPU mode, just takes a little longer.
…On Wed, 10 Jan 2024, 4:52 am COOLak, ***@***.***> wrote:
Do you know for sure that it will finish properly if I buy some credits? I
can't figure if it means uninterrupted work or it can still throw a captcha
mid-process. Or I should buy a VM from them and work on it. The latter is
too technical for me though, haha. On the other hand, don't wanna throw
money in vain.
—
Reply to this email directly, view it on GitHub
<#444 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AB3UVPO2TGTXFGQTFFHLDHDYNV7XRAVCNFSM6AAAAAAVUKH3CCVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4DANZQGE2DA>
.
You are receiving this because you commented.Message ID:
<Anjok07/ultimatevocalremovergui/repo-discussions/444/comments/8070140@
github.com>
|
|
ive used the settings recommded, but my instrumental track is muffled.. any ideas what i can do? |
|
If anyone's interested in keeping not only instrumentals but also the backup vocals/harmonies, I've used ensemble mode with VR Arc: 5_HP-Karaoke-UVR and VR Arc: 6_HP-Karaoke-UVR. They were titled a little differently when downloading, but they were the only two in the VR Arch dropdown with Karaoke in the title. There are Karaoke downloads in the MDX-Net dropdown but I haven't tried them yet. |
|
What's the difference between 5_HP-Karaoke-UVR and 6_HP-Karaoke-UVR? I can't find any information on them and their differences, I'm trying to extract the best possible backing vocals with them |
|
Thanks DUDE IT IS ABSOLUTELY THE SHIT!!! Worked insanely great for a Chinese song. Now all i need is to gently fix some of the vocal samples in the instrumental by simply dragging them and assemble to the AI outputs. Thanks alot, really. 🤗 |
|
Here's my setup with no backup vocal.
Overlap Mdx- 0.25 I've tried using VR models but the audio seems muddy so I skipped it. This is not a setting for extracting each instrument. UPDATE Ensemble Algo - Min/Min -Mdx settings- |
Hello man, I'm trying to follow your specs but I'm confused, it produced 4 instrumental files, is that expected? I'm expecting just one instrumental combined/ processed by 4 models. Here's screenshot of my main window: Can you tell me what I'm doing wrong? Thank you. |
Did you mean there's 4 output ? Try disable "Save all output" in ensemble settings. If that's what you meant. |
nice much better! thanks man! these settings actually sound good compared to other i've tried, what's your settings for acappela or vocal only im curious |
|
Actually, for an optimal result you should use the Kim Duality Version 2 Roformer model in MDX mode, without using ensemble |
|
the Kim Duality Version 2 Roformer model? where, there's nothing like that, |
|
Hi there! For the RoFormer version, you need to download the new package. You can find it directly via this link: https://github.com/Anjok07/ultimatevocalremovergui/releases/download/v5.6/UVR_1_15_25_22_30_BETA_full.exe Once you have installed the new version of UVR, go to the download section and download the Dualty version 2 package. |
|
Would it work, though? Every model in UVR5 I tried, and even other tools like UVR5, and I still have the same issues: the vocals are muffled and bleeding into the instrument parts |
|
Have you tried putting the separated tracks into a DAW (like FL Studio, Ableton, or Logic)?Since the AI extraction can sometimes leave the tracks sounding a bit dull, you actually need to post-process them. Try adding an EQ to boost the high frequencies (around 4kHz - 8kHz) to fix the muffled vocal sound. Then, you can use a Compressor and a Gate on the instrumental track to squash down and hide the remaining vocal bleeding.Give it a shot inside a DAW, it usually cleans up the results drastically! |
|
I'm trying to extract instruments in different tracks, I need to make some arrangements to a song, and I need to separate instruments, so i can edit for example drums, and have control of editing bass, vocals and harmonies. Does anyone have used a successful processing model to achieve this? |
and others
MDX-NET UVR-MDX-NET Inst 3 but if you want to get highest quality result you should try using ensemble mode instead (combine results of multiple models together)