The program is a shell for neural network models designed to work with a mix and extract audio tracks from it. The program extracts vocals from the mix, and also divides the mix into tracks (drum section, bass, keys, vocals, guitar, etc.). Multi-track splitting is available on Demucs models!

Three powerful algorithms are presented:
Demucs
MDX-Net
VR
Each of the algorithms has its own unique capabilities and models of neural networks.
To extract four or six tracks, the DEMUCS algorithm is used , which consists of several models:
htdemucs - The latest, fourth and most powerful AI model
hdemucs_mmi , htdemucs_ft - Versions of the fourth Demucs. Trained on the basis of MUSDB HQ.
htdemucs_6s - AI model for extracting six tracks.
Models are also presented:
mdx
mdx_extra
mdx_extra_q
mdx_q
UVR_Demucs_Model

, etc.
It should be noted that the mdx_extra_q model is most often found on paid sites for dividing a mix into 4 tracks.
VR MODELS - AI models that were presented at the Sony Music Demixing Challenge and trained at MUSDB HQ. Ideal for separating vocals and backing tracks, but lose a bit of hdemucs in some cases where the mix has problems with phase and a too wide SIDE. These models are perfect for getting vocals out of a mix. There are several models, they work in different ways:
10_SP-UVR-2B-32000-1
11_SP-UVR-2B-32000-2
12_SP-UVR-3B-44100
13_SP-UVR-4B-44100-1
14_SP-UVR-4B-44100-2
15_SP-UVR-MID-44100-1
16_SP-UVR-MID-44100-2
1_HP-UVR
2_HP-UVR
3_HP-Vocal-UVR
4_HP-Vocal-UVR
5_HP-Karaoke-UVR
6_HP-Karaoke-UVR
7_HP2-UVR
8_HP2-UVR
9_HP2-UVR
baseline
HighPrecison_4band_1
HighPrecison_4band_2
LOFI_2band-2_33966KB
MGM-v5-KAROKEE-32000-BETA1
MGM-v5-KAROKEE-32000-BETA2-AGR
MGM-v5-Vocal_2Band-32000-BETA1
MGM_HIGHEND_v4
MGM_LOWEND_A_v4
MGM_LOWEND_B_v4
MGM_MAIN_v4
Multi-Genre Model
NewLayer_4band_1
NewLayer_4band_3
NewLayer_4band_arch-130m_2
Stacked Model1
Stacked Model2
StackedMGM_1band_arch-default
Stock Model
MDX-Ne t Models Work well with old vinyl recordings, etc., but take longer to process. They do an excellent job in tracks where there are synthesized sounds that are timbre similar to vocals. There are several models, all of them have different algorithms. The kuielab models aim to extract a specific, single track:
kuielab_a_bass
kuielab_a_drums
kuielab_a_other
kuielab_a_vocals
kuielab_b_bass
kuielab_b_drums
kuielab_b_other
kuielab_b_vocals
UVR-MDX-NET-Inst_1
UVR-MDX-NET-Inst_2
UVR-MDX-NET-Inst_3
UVR-MDX-NET-Inst_Main
UVR_MDXNET_1_9703
UVR_MDXNET_2_9682
UVR_MDXNET_3_9662
UVR_MDXNET_9482
UVR_MDXNET_KARA
UVR_MDXNET_Main
• VR Architecture - These models use spectrograms to separate sources. In fact, this is a simplified Ensemble Mode method
• MDX-Net - Hybrid between models
• Demucs - Allows you to extract up to 6 tracks. Hit'n'Mix RipX uses the oldest version of this model, which runs on the CPU.
• Ensemble Mode - A hybrid of several models. You can configure it yourself, so that the extracted tracks would be additionally processed by other models.
• Audio Tools - several additions to work with pitch and track time.



IMPORTANT! To work correctly with the GPU, it must have at least 3 GB V-RAM, 6 is recommended.
Please note that the new version differs from the old one not only in interface, but also in functionality. Hints are enabled in the settings.

Installation
Place the folder from the image on DISK C
Placing the folder on other disks will cause the application to become unstable!
The guide for using the application is located in the Manual
folder The application is launched using the UVR.exe file

Disgusting: 64 bit
Interface language: English
System requirements: Windows 10, 11 64bit


What News?
Official site does not provide any info about changes in this version.

https://github.com/Anjok07/ultimatevocalremovergui